Advancing image segmentation with DBO-Otsu: Addressing rubber tree diseases through enhanced threshold techniques.

1 Introduction

Addressing the profound impact of Tapping Panel Dryness (TPD) on yield and quality in the global rubber industry, this study introduces a cutting-edge Otsu threshold segmentation technique, enhanced by Dung Beetle Optimization (DBO-Otsu). This innovative approach optimizes the segmentation threshold combination by accelerating convergence and diversifying search methodologies. Following initial segmentation, TPD severity levels are meticulously assessed using morphological characteristics, enabling precise determination of optimal thresholds for final segmentation. The efficacy of DBO-Otsu is rigorously evaluated against mainstream benchmarks like Peak Signal-to-Noise Ratio (PSNR), Structural Similarity Index (SSIM), and Feature Similarity Index (FSIM), and compared with six contemporary swarm intelligence algorithms. The findings reveal that DBO-Otsu substantially surpasses its counterparts in image segmentation quality and processing speed. Further empirical analysis on a dataset comprising TPD cases from level 1 to 5 underscores the algorithm's practical utility, achieving an impressive 80% accuracy in severity level identification and underscoring its potential for TPD image segmentation and recognition tasks.

1.1 Rubber tree tapping panel dryness

The rubber tree, a pivotal economic crop, significantly contributes to the global economy with its primary product, natural rubber [[1]]. Growth, latex yield, and health of rubber trees are vulnerable to various environmental and human-induced factors [[2]–[4]]. Tapping Panel Dryness (TPD) poses a major challenge during latex extraction, leading to latex tube degeneration and considerable yield reduction [[3], [5]–[8]]. Unraveling the mechanisms underlying TPD is essential for boosting resilience and latex production in rubber trees and for the sustainable evolution of the rubber industry [[9]]. Recent advancements in the molecular understanding of TPD [[2], [7], [9], [11]] have been significant, yet their practical application in disease identification remains hindered by traditional methods. The industry's current reliance on visual inspections by experienced professionals is limited by subjectivity and varying efficiency and accuracy levels. With extensive global rubber plantations and diverse disease manifestations, traditional manual methods fall short. Image processing technologies emerge as a superior alternative, offering objectivity and robust big data handling, thus providing precise disease assessments essential for research and developing effective disease management strategies in rubber trees.

The fundamental goal in image recognition, especially considering the unique image characteristics of the cuts and latex, is to achieve precise segmentation. High-quality image segmentation directly contributes to enhanced diagnostic accuracy, promoting objectivity and uniformity in evaluations. This approach facilitates automatic disease identification in TPD-affected trees across different stages, assisting researchers in real-time monitoring and future latex yield prediction, and propelling forward the study of rubber tree diseases.

1.2 Image segmentation

Image segmentation, an integral component in image processing, lays the technical groundwork for condition diagnosis by isolating image regions with varying characteristics [[13]]. While traditional segmentation methods primarily leverage threshold setting, histogram analysis [[14]], region growing, fuzzy clustering [[15]], K-means clustering [[17]], and edge detection [[18]], advanced techniques incorporate active contours, graph cuts, and sophisticated mathematical and probabilistic models [[20]]. Notably, deep learning approaches [[21]–[25]] like Fully Convolutional Networks (FCN) [[26]],U-Net [[27]],PSPNet [[28]] and FC-DenseNet have revolutionized segmentation with their high precision in pixel-level classification. Deep learning methods offer unmatched segmentation accuracy and efficiency; however, automated threshold segmentation techniques remain popular for their simplicity and effectiveness [[29]]. For example, the multi-threshold Tsallis entropy recursive algorithm by Wang et al. [[31]] accelerates segmentation while ensuring efficiency. Sharma et al. [[32]] introduced an optimized multi-level threshold segmentation algorithm, proving its efficacy in brain tumor segmentation and advancing threshold segmentation research. Lei et al. [[33]] proposed an adaptive granularity Renyi rough entropy method, which augments threshold segmentation accuracy and speed, demonstrating its utility in rapid and efficient image segmentation.

In the context of TPD in rubber plantations, the environmental complexity and signal instability demand more timely and robust recognition technologies. Despite deep learning's superior performance in image segmentation, its high hardware requisites restrict application in wearable devices. Threshold-based segmentation, known for its lightweight and efficient nature, becomes a fitting alternative, especially suitable for wearable device integration, offering vital support for intelligent TPD recognition. Therefore, advancing research and development of these algorithms is imperative for managing rubber tree diseases and facilitating early diagnosis.

1.3 Application of Otsu algorithm in image segmentation

The Otsu algorithm, serving as an adaptive threshold segmentation method, has proven highly effective in images with bimodal histograms [[34]]. Despite its widespread adoption, the algorithm encounters performance limitations in scenarios with highly variable background and target intensity, or significant noise disturbances. For instance, its robustness is compromised in high-noise images affected by salt-and-pepper noise, leading to segmentation inaccuracies [[35]]. To address these challenges, novel improvements have been proposed. Notably, the integration of the 3D Otsu algorithm with local contrast enhancement has significantly ameliorated segmentation quality while preserving edge details [[36]]. Methods combining pixel intensity with spatial context, through energy curve optimization, have shown promise under varying lighting conditions, yet they grapple with dynamic environments [[37]]. Hybrid algorithms, like the amalgamation of Otsu with K-means clustering, have enhanced accuracy in multi-light spot center detection but impose greater computational demands [[38]]. Additionally, the 2D Otsu algorithm, when coupled with adaptive energy segmentation and genetic algorithms, demonstrates efficiency, albeit with lingering challenges in handling complex textures and color variations [[39]].

In this research domain, the application of metaheuristic optimization algorithms is crucial. These algorithms, inspired by natural phenomena and artificial intelligence, are adept at tackling diverse and intricate optimization challenges [[41]]. Algorithms such as genetic algorithms, Whale Optimization Algorithm (WOA), Particle Swarm Optimization (PSO), and Harris Hawks Optimization (HHO) have each contributed uniquely to threshold selection, each with distinct strengths and weaknesses [[43]–[54]]. Recent advancements include swarm intelligence algorithms for multi-threshold segmentation, particularly effective in processing COVID-19 chest X-rays and CT scans [[55]–[59]]. Chen et al. [[59]] augmented the Artificial Bee Colony algorithm with dynamic strategies, boosting initial convergence and global search efficiency. Abualigah et al. [[56]] innovated with a multi-threshold method based on the Arithmetic Optimization Algorithm (AOA), DAOA, enhancing local search capabilities through differential evolution techniques. Liu et al. [[55]] merged ant colony optimization with Cauchy mutation and Levy flight strategies, significantly elevating search efficiency and segmentation precision. Emam et al. [[60]] devised an enhanced Reptile Search Algorithm (mRSA) optimizing both global optimization and image segmentation, showcasing remarkable performance in MRI brain image multi-threshold segmentation. Chen et al. [[61]] introduced the HVSFLA algorithm, ensuring diverse and active search mechanisms, excelling in multi-threshold segmentation applications for invasive ductal carcinoma of the breast. Abdel-Basset et al. [[62]] proposed an improved balance optimization algorithm for optimal threshold discovery in grayscale images. These advancements not only propel swarm intelligence applications in medical image processing but also offer potent tools for medical decision-making.

On the other hand, the integration of metaheuristic algorithms with the Otsu method has significantly advanced its capabilities. A study by [[63]] introduced the DE-GWO-Otsu algorithm, a hybrid of Differential Evolution (DE), Grey Wolf Optimization (GWO), and Otsu's method. This approach addressed the stability and local optima challenges of the GWO. In another innovation, [[64]] proposed the FOA-Otsu method, merging the Fruit Fly Optimization Algorithm with the Otsu technique, which considerably enhanced real-time image segmentation performance while halving segmentation time. Additionally, [[65]] developed an Improved Golden Jackal Optimization algorithm (IGJO) integrated with the Otsu method, markedly boosting the accuracy and efficiency in skin cancer image segmentation. The use of the AOA by [[66]] for determining optimal thresholds in multi-layer segmentation demonstrated effectiveness when coupled with the Otsu fitness function. Furthermore, Rather et al. [[67]] employed a Levy flight and chaos theory-based Gravity Search Algorithm (LCGSA) to optimize computational efficiency in multi-threshold segmentation, overcoming traditional segmentation issues like local minima and premature convergence. Liu et al. [[68]] innovated with the HCROA, a primate-inspired WOA, combined with the Chimp Optimization Algorithm, to enhance exploration and exploitation balance, thereby improving segmentation accuracy and noise robustness. Finally, [[69]] merged Enhanced Fuzzy Elephant Herd Optimization (EFEHO) with the Otsu method, facilitating rapid diagnosis in Alzheimer's disease and Mild Cognitive Impairment (MCI) contexts.

Despite the significant progress made by metaheuristic algorithm-enhanced Otsu methods in various application domains, their robustness [[70]] and segmentation accuracy remain inadequate when dealing with images containing complex lighting, angles, and texture variations, such as rubber tree tapping scars and latex images. The computational complexity is also relatively high. To address this challenge, this study introduces the DBO-Otsu algorithm, a novel integration of the classic Otsu's method with the innovative Dung Beetle Optimizer (DBO), specifically targeting complex image segmentation tasks. The DBO algorithm, inspired by the natural behavior of dung beetles, such as their unique rolling and foraging strategies, effectively enhances search diversity, efficiency, and global convergence [[71]]. Compared to other metaheuristic algorithms, DBO exhibits pronounced advantages in multi-threshold image segmentation tasks, particularly in terms of convergence speed and solution precision, which are crucial for accurately segmenting TPD in this study. Moreover, the effectiveness of this algorithm has been proven in various practical applications: DBO has demonstrated significant performance improvements in spaceborne SAR image waterbody detection [[73]], lung cancer detection and classification [[75]], and pesticide residue identification in rapeseed oil [[76]]. These successful cases further validate our choice of DBO as the framework for improvement. Additionally, according to the No Free Lunch (NFL) theorem [[77]], no optimization algorithm excels in all problem types, thus spurring the development of new algorithms and the enhancement of existing ones. Therefore, selecting DBO as the optimization algorithm for this study is based on its unique strengths in addressing specific categories of problems. The advantages of the algorithm are illustrated in Fig 1.

Graph: Fig 1 info:doi/10.1371/journal.pone.0297284.g001

This paper's primary contributions are as follows: (1) The development of the DBO-Otsu algorithm, tailored specifically for complex image segmentation challenges, markedly improving processing efficiency and accuracy. (2) An innovative enhancement of the traditional Otsu method within the DBO-Otsu framework, involving an initial preprocessing stage for multi-threshold segmentation to remove low gray-scale areas, thereby focusing on high gray-scale segments, particularly latex and scars. (3) A comprehensive evaluation of the DBO-Otsu algorithm through a suite of established performance metrics, showcasing its superior performance across various dimensions. (4) An in-depth exploration of the DBO algorithm's application potential in image segmentation, substantiated by practical use cases, notably in diagnosing rubber tree TPD. The paper is organized into subsequent sections as follows: Section 2 elucidates the principles of the Otsu algorithm and the workings of the DBO mechanism; Section 3 elaborates on the DBO-Otsu algorithm's implementation and its innovative aspects relative to the conventional Otsu method; Section 4 demonstrates the algorithm's effectiveness and comparative analysis via experimental results; Section 5 concludes with a summary of the findings and a discussion on prospective applications.

2 Theoretical foundations

2.1 Otsu algorithm

The Otsu algorithm is a technique for image binarization segmentation based on global adaptive thresholding [[78]]. Its core idea revolves around selecting the optimal threshold by calculating the maximum inter-class variance using the gray level histogram of the image [[79]] Let's consider a digital image of size M × N, containing L distinct gray levels, represented as the set {0, 1, 2, ..., L − 1}. If ni denotes the number of pixels at gray level i, then the image's total pixel count is represented as MN = n0 + n1 + n2 + ... + nL−1. Consequently, the normalized histogram is defined by the ratio of the pixel count for each gray level to the total pixel count, pi = ni/MN, from which we have

Graph

$$\begin{array}{c}\hfill \sum _{i=0}^{L-1}{p}_i=1,p_i\ge 0\end{array}$$ (1)

Consider the threshold T(k) = k where 0 < k < L − 1. The input image is categorized into two classes: C1 and C2.C1 encompasses pixels with gray values in the range [0, k] while C2 includes those in the range [k + 1, L − 1]. Based on this, the probabilities P1(k) and P2(k) represent classifications into C1 and C2 respectively.

Graph

$$\begin{array}{c}\hfill P\left(k\right)=\{\begin{array}{cc}{P}_1\left(k\right)={\displaystyle \sum _{j=0}^k}{p}_j\hfill & \\ P_2\left(k\right)={\displaystyle \sum _{j=k+1}^{L-1}}{p}_j=1-P_1\left(k\right)\hfill \end{array}\end{array}$$ (2)

For a given threshold value,T(k), we denote the average gray value of pixels in class C1 as m1(k) and in class C2 as m2(k). Respectively:

Graph

$$\begin{array}{c}\hfill \{\begin{array}{c}{m}_1\left(k\right)=\frac{1}{P_1\left(k\right)}{\displaystyle \sum _{i=0}^k}i{p}_i\hfill \\ m_2\left(k\right)=\frac{1}{P_2\left(k\right)}{\displaystyle \sum _{i=k+1}^{L-1}}i{p}_i\hfill \end{array}\end{array}$$ (3)

The average gray level of the image is defined as:

Graph

$$\begin{array}{c}\hfill m_G={\displaystyle \sum _{i=0}^{L-1}}i{p}_i\end{array}$$ (4)

Graph

${\sigma }_B^2$ represents the between-class variance, with its formula being:

Graph

$$\begin{array}{c}\hfill {\sigma }_B^2=P_1{(m_1-m_G)}^2+P_2{(m_2-m_G)}^2\end{array}$$ (5)

Again citing k, the end result is:

Graph

$$\begin{array}{c}\hfill {\sigma }_B^2\left(k\right)=\frac{{[m_G{P}_1\left(k\right)-m\left(k\right)]}^2}{P_1\left(k\right)[1-P_1\left(k\right)]}\end{array}$$ (6)

Thus the optimal threhold is k*, which maximizes

Graph

${\sigma }_B^2\left(k\right)$ :

Graph

$$\begin{array}{c}\hfill {\sigma }_B^2\left(k^{*}\right)=\underset{0\le k\le L-1}{\text{max}}{\sigma }_B^2\left(k\right)\end{array}$$ (7)

For multilevel threshold segmentation, the assumption is that m threshold levels (t1, t2, ..., tm) segment the image into m + 1 categories: C0, C1, C2, ..., Cm. The objective function for the segmentation process is:

Graph

$$\begin{array}{c}\hfill \begin{array}{cc}\hfill J{\left(t\right)}_{max}=& {\sigma }_0+{\sigma }_1+\cdots +{\sigma }_m\hfill \\ \hfill {\sigma }_0=& {\omega }_0{(m_0-m_G)}^2\hfill \\ \hfill {\sigma }_1=& {\omega }_1{(m_1-m_G)}^2\hfill \\ & \cdots \hfill \\ \hfill {\sigma }_m=& {\omega }_m{(m_m-m_G)}^2\hfill \end{array}\end{array}$$ (8)

From the analysis of the outlined objective function, it becomes clear that the algorithm's solution space extends over a q − 1 dimensional realm, with q indicating the total count of thresholds. Within this multidimensional space, specific calculations are crucial, primarily those centered around the inter-class variance, which include determining averages among different classes. Considering these computations, the time complexity for executing multi-level threshold segmentation as per the Otsu method escalates to O(Lq), where L signifies the quantity of gray scale levels. The exhaustive nature of computations across the q − 1 dimensional space results in an exponential surge in time complexity relative to the increase in threshold numbers. Contrasting with single-level threshold techniques, multi-level threshold approaches adopt a greater number of thresholds, thereby capturing a more detailed essence of the image. Consequently, while multi-level threshold segmentation furnishes enhanced image detail, it simultaneously amplifies computational complexity. Striking an optimal balance between computational time and segmentation accuracy is imperative, thereby mandating the selection of an apt number of thresholds for effective image segmentation.

2.2 DBO algorithm

The position update of the beetle during its rolling behavior can be characterized using a specific mathematical model:

Graph

$$\begin{array}{c}\hfill \begin{array}{cc}\hfill x_i(t+1)=x_i& (t)+\alpha \times k\times x_i(t-1)+b\times \Delta x\hfill \\ & \Delta x=|x_i\left(t\right)-X^{\omega }|\hfill \end{array}\end{array}$$ (9)

Let t represent the current iteration number, serving to control the algorithm's iterative process. The symbol xi(t) denotes the position of the ith dung beetle at the tth iteration, signifying a candidate solution in the solution space. Additionally, k ∈ (0, 0.2] represents a deflection coefficient constant, essential for controlling the dung beetle's deflection degree during its search. Another constant, denoted by b, belongs to the range (0, 1), where α is a specific coefficient with values of either -1 or 1 (refer to Algorithm 1). Lastly, Xω signifies the global worst position, whereas Δx models the changes in light intensity.

Algorithm 1 Selection strategy for a

Input:

probability value l

Output:

natural coefficients a

h ← rand(1)

if h > lthen

 a ← 1

else

 a ← −1

end if

When a dung beetle encounters an obstacle during its rolling phase and is hindered from proceeding, it resorts to a reorientation dance to identify a new direction. Consequently, the position during this dancing behavior is defined by:

Graph

$$\begin{array}{c}\hfill x_i(t+1)=x_i\left(t\right)+\text{tan}\left(\theta \right)|x_i\left(t\right)-x_i(t-1)|\end{array}$$ (10)

where θ ∈ [0, π], if θ is equal to 0, neither

Graph

$\frac{\pi }{2}$ nor π will update the dung beetle' s position.

In dung beetle optimization algorithms, the choice of apt spawning sites by female dung beetles plays a pivotal role in ensuring the survival and procreation of their progeny. To model the behavior of female dung beetles when selecting a spawning area, we employ a boundary selection strategy as follows:

Graph

$$\begin{array}{c}\hfill \begin{array}{cc}\hfill L{b}^{*}& =\text{max}(X^{*}\times (1-R),Lb)\hfill \\ \hfill U{b}^{*}& =\text{min}(X^{*}\times (1-R),Ub)\hfill \end{array}\end{array}$$ (11)

here, X* represents the current local optimal position. The symbols Lb*, and Ub* define the lower and upper boundaries of the spawning area, respectively. Furthermore, R = 1 − t/Tmax, Tmax are maximum iteration numbers, while Lb and Ub specify the lower and upper constraints of the optimization problem. In the Dung Beetle Optimization Algorithm (DBO), upon establishing the spawning area, female dung beetles prioritize breeding balls within that vicinity for laying eggs. It's pivotal to highlight that every female dung beetle within the DBO framework produces a single breeding ball per iteration. The position of these breeding balls remains fluid throughout the iteration process, represented as:

Graph

$$\begin{array}{c}\hfill B_i(t+1)=X^{*}+b_1\times (B_i\left(t\right)-L{b}^{*})+b_2\times (B_i\left(t\right)-U{b}^{*})\end{array}$$ (12)

in this context, the position of the ith breeding ball during the tth iteration is symbolized by Bi(t), with b1 and b2 serving as two distinct random vectors, each of size 1×D. Here, D encapsulates the optimization problem's dimensionality. Importantly, the positioning of breeding balls adheres strictly to the confines of the designated spawning area. (Refer to Algorithm 2 for further details.)

Algorithm 2 Breeding ball position update strategy

Input:

maximum number of iterations Tmax, number of breeding balls N,current number of iterations t

Output:

Location of the ith breeding ball Bi

R = 1 − t/Tmax

for i ← 1 tondo

 Update the position of the breeding ball using Eq (12)

 forj ← 1 toDdo

  ifBij > Ub* then

   Bij ← Ub*

  end if

  ifBij < Lb* then

   Bij ← Lb*

  end if

 end for

end for

Fig 2 depicts the movement of rolling dung beetles, represented by dark blue dots, in a three-dimensional search space. The yellow dot at the center of a small sphere indicates the current local optimal position, X*, representing the best solution found in the current iteration. Within this sphere, small black dots symbolize breeding balls, each enclosing a dung beetle egg. Red dots at the extremities of both the large and small spheres demarcate the upper and lower boundary limits. These boundaries restrict the beetles' rolling and egg-laying range, ensuring they search and reproduce effectively within the algorithm's optimal range.

Graph: Fig 2 info:doi/10.1371/journal.pone.0297284.g002

Adult dung beetles, often referred to as 'baby dung beetles', emerge from the ground in search of food. To model the foraging behavior of dung beetles in their natural habitat, it's essential to define an optimal foraging area. The boundaries of this area are delineated as follows:

Graph

$$\begin{array}{c}\hfill \begin{array}{cc}\hfill L{b}^b& =\text{max}(X^b\times (1-R),Lb)\hfill \\ \hfill U{b}^b& =\text{min}(X^b\times (1-R),Ub)\hfill \end{array}\end{array}$$ (13)

In this, Xb represents the global optimal position while Lbb and Ubb respectively indicate the lower and upper bounds of the optimal foraging area. Further parameter definitions are given in Eq (11). Consequently, the position of the dung beetle is updated as:

Graph

$$\begin{array}{c}\hfill x_i(t+1)=x_i\left(t\right)+C_1\times (x_i\left(t\right)-L{b}^b)+C_2\times (x_i\left(t\right)-U{b}^b)\end{array}$$ (14)

here, xi(t) specifies the position of the ith dung beetle during the tth iteration. This update process involves two random vectors:C1 and C2. The former, C1, is a random number following a normal distribution, aiding in modulating the exploratory behavior of the dung beetle. Meanwhile,C2 is a random vector within the interval (0, 1), adjusting the beetle's position in relation to both the globally optimal position and the optimal foraging area.

Additionally, a category of dung beetles, termed 'thieves', is integrated into the algorithm. Their primary role is to pilfer dung balls from fellow beetles for sustenance. As inferred from Eq (13), Xb symbolizes the prime food source. It's plausible, then, to consider the vicinity of Xb as the prime zone for food competition. As iterations proceed, the position data of these thieving dung beetles evolves and is characterized as follows:

Graph

$$\begin{array}{c}\hfill x_i(t+1)=X^b+S\times g\times \left(\right|x_i\left(t\right)-X^{*}|+|x_i\left(t\right)-X^b\left|\right)\end{array}$$ (15)

in this representation, xi(t) indicates the position of the ith thieving dung beetle at the tth iteration. Additionally, g is a random vector of dimensions 1×D, adhering to a normal distribution, and S is a constant.

Building upon the preceding discussion, the devised DBO algorithm first determines the maximum iteration count and sets the total population size of dung beetles as N. All agents are subsequently initialized at random, with their roles distributed based on a specified proportionate diagram. This distribution is visualized with sectors, where 20% corresponds to ball-rolling dung beetles, 20% to ball-breeding dung beetles, 25% to small dung beetles, and the remaining 35% to stealing dung beetles.

For illustrative purposes, let's assume a total population of 30 dung beetles. Using Fig 3 as a guide, beetles are allocated to each agent category. Here, orange, yellow, green, and brown rectangles symbolize rolling dung beetles, breeding balls, small dung beetles, and stealing dung beetles respectively. This allocation ensures that during the algorithm's operation, dung beetles of distinct roles synergize based on their unique behaviors, aiming for enhanced optimization.

Graph: Fig 3 info:doi/10.1371/journal.pone.0297284.g003

Subsequently, the positions of the rolling dung beetle, breeding ball, little dung beetle, and stealing dung beetle are incessantly refreshed. Guided by specific rules and equations within the algorithm, they undergo adaptive shifts through iterative processes. Ultimately, the algorithm presents the global best position Xb, accompanied by its respective fitness value.

3 DBO-Otsu

3.1 Improvement of the traditional Otsu algorithm

A rubber cut mark map is randomly taken as in Fig 4, and its grayscale histogram is shown in Fig 5. In this experiment, the threshold value of the latex and cut mark region of interest is between 150–255, and it can be seen from the histogram that the traditional Otsu algorithm is affected by the global pixel distribution, and the threshold value (shown by the red solid line in the figure) will be to the left, which is affected by a large number of low grayscale regions, and it is unable to segment the region of our interest. If this threshold is used for segmentation, the segmentation map is shown in Fig 6, and it is obvious that it is impossible to distinguish the cut marks from the latex.

Graph: Fig 4 info:doi/10.1371/journal.pone.0297284.g004

Graph: Fig 5 info:doi/10.1371/journal.pone.0297284.g005

Graph: Fig 6 info:doi/10.1371/journal.pone.0297284.g006

After experimental comparisons, the following improvements are proposed. The image of interest is represented by L gray levels (1; 2; ...; L). First, a suitable gray scale Th is set as the first threshold. According to the selected threshold, the gray scale of the image is divided into two parts:[0, Th] and [Th + 1, L − 1]. For an image with pixels N × M, the number of pixels with gray level i is n, and the total number of pixels n is

Graph

$$\begin{array}{c}\hfill n=M\times N={\displaystyle \sum _{i=0}^{L-1}}{n}_i\end{array}$$ (16)

The number of pixels in [0; Th] is nl, and the number of pixels in [Th + 1; L − 1] is nr.

Graph

$$\begin{array}{c}\hfill n_l={\displaystyle \sum _{i=0}^{Th}}{n}_i\text{and}{n}_r={\displaystyle \sum _{i=Th+1}^{L-1}}{n}_i\end{array}$$ (17)

The probability that a pixel is in [0; Th] is pil and the probability that a pixel is in [Th + 1; L − 1] is pir.

Graph

$$\begin{array}{c}\hfill p_{il}=\frac{n_i}{n_l}\text{and}{p}_{ir}=\frac{n_i}{n_r}\end{array}$$ (18)

Setting the gray values as j, k, l, m, the range of [Th + 1; L − 1] is divided into five categories: C0,C1,C2,C3 and C4. The distribution probability of C0,C1,C2,C3 and C4 is ω0, ω1, ω2, ω3 and ω4, denoted as:

Graph

$$\begin{array}{c}\hfill \begin{array}{cc}& {\omega }_0={\displaystyle \sum _{i=Th+1}^j}{p}_{ir}\text{and}{\omega }_1={\displaystyle \sum _{i=j+1}^k}{p}_{ir}\hfill \\ & {\omega }_2={\displaystyle \sum _{i=k+1}^l}{p}_{ir}\text{and}{\omega }_3={\displaystyle \sum _{i=l+1}^m}{p}_{ir}\hfill \\ & {\omega }_4={\displaystyle \sum _{i=m+1}^{L-1}}{p}_{ir}\hfill \end{array}\end{array}$$ (19)

The average pixel gray probabilities of C0,C1,C2,C3 and C4 are μ0, μ1,μ2, μ3 and μ4.

Graph

$$\begin{array}{c}\hfill \begin{array}{cc}& {\mu }_0={\displaystyle \sum _{i=Th+1}^j}i·p_{ir}\text{and}{\mu }_1={\displaystyle \sum _{i=j+1}^k}i·p_{ir}\hfill \\ & {\mu }_2={\displaystyle \sum _{i=k+1}^l}i·p_{ir}\text{and}{\mu }_3={\displaystyle \sum _{i=l+1}^m}i·p_{ir}\hfill \\ & {\mu }_4={\displaystyle \sum _{i=m+1}^{L-1}}{p}_{ir}\hfill \end{array}\end{array}$$ (20)

The average gray level μ in the range [Th + 1; L − 1] can be expressed as follows:

Graph

$$\begin{array}{c}\hfill \mu ={\displaystyle \sum _{i=Th+1}^{L-1}}i·p_{ir}\end{array}$$ (21)

The between-class variances for C0,C1,C2 and C3 were:

Graph

$$\begin{array}{c}\hfill {\sigma }_B^2={\omega }_0{({\mu }_0-\mu )}^2+{\omega }_1{({\mu }_1-\mu )}^2\end{array}$$ (22)

Referring to j, k, l, m, the optimal threshold are j*, k*, l*, m* such that the maximum value is reached

Graph

${\sigma }_B^2(j^{*},k^{*},l^{*},m^{*})$

Graph

$$\begin{array}{c}\hfill {\sigma }_B^2(j^{*},k^{*},l^{*},m^{*})=\underset{1\le j<k<l\le L-1}{\text{max}}{\sigma }_B^2(j,k,l,m)\end{array}$$ (23)

The obtained j*, k*, l*, m* range is in [Th + 1; L − 1]. The improved version processes [Th + 1; L − 1] as separate images. As a result, the effect of a large number of pixels in the low gray range on the region of interest can be ignored in the selection of the threshold.

3.2 Otsu method improved with DBO algorithm

The time complexity of the improved Otsu method is O(L4). In order to reduce the computation time, we combine the DBO algorithm with the improved Otsu method and propose the DBO-Otsu method. The specific steps are as follows, as illustrated in Fig 7.

Graph: Fig 7 info:doi/10.1371/journal.pone.0297284.g007

In the DBO-Otsu algorithm, the gray scale value K during Otsu calculation is considered as the coordinates X of the dung beetle population in the algorithm, and according to Eq (23) The fitness of each dung beetle individual is calculated, and the fitness is inverted. Then the DBO algorithm is used to simulate the behavioral patterns of dung beetles, comparing the fitness values and updating the coordinates X in the iterative process, and finally finding the optimal threshold value to replace the exhaustive method in the traditional Otsu algorithm.

It should be noted that due to the design of the algorithm, the coordinates obtained result in floating point numbers, while the grayscale values of the image are in the discrete integer range [0, 255]. Therefore, when calculating the individual fitness, the coordinates need to be processed and limited to integers for subsequent calculations.

The DBO-Otsu pseudocode is shown in Algorithm 3.

Algorithm 3 DBO-Otsu algorithm

Inputs: maximum number of iterations Tmax, population size N

Outputs: optimal position Xb and its fitness value fb

Randomly initialize the dung beetle population i ← 1, 2, ..., N

Initialize parameters: Dim = 4, bounds ∈ [1, 256], T = 150, t = 0, N = 60

while t ≤ Tmax

 fori ← 1 toNdo

  ifi = = Dung Beetle then

   δ = rand(1)

   ifδ < 0.9 then

    Use Algorithm 1 to select α

    Update location using formula (9)

   else

    Update location using Eq (12)

   end if

  else ifi = = Breeding Balls then

   Update using Algorithm 2

  else ifi == Little Dung Beetle then

   Update using Eq (14)

  else ifi == Stealing Dung Beetles then

   Update using formula (15)

  end if

 end for

 if new position is better then

  Update it

 end if

 t = t + 1

end while

return Adaptation value fb

3.3 Segmentation strategies of DBO-Otsu at different levels

In low-level Tapping Panel Dryness (TPD) images, where the latex quantity has not significantly diminished, as illustrated in Figs 8–10.

Graph: Fig 8 info:doi/10.1371/journal.pone.0297284.g008

Graph: Fig 9 info:doi/10.1371/journal.pone.0297284.g009

Graph: Fig 10 info:doi/10.1371/journal.pone.0297284.g010

The final threshold in the multi-threshold output of the DBO-Otsu algorithm provides a high-quality segmentation of the latex. However, for images of higher-level TPD, where latex is sparse, as shown in Figs 11–13, the original approach, resulting in Figs 14–16, often fails to reflect the actual scenario. In these cases, the latex regions no longer manifest as distinct peaks on the grayscale histogram, rendering traditional multi-threshold segmentation methods ineffective in isolating the latex areas.

Graph: Fig 11 info:doi/10.1371/journal.pone.0297284.g011

Graph: Fig 12 info:doi/10.1371/journal.pone.0297284.g012

Graph: Fig 13 info:doi/10.1371/journal.pone.0297284.g013

Graph: Fig 14 info:doi/10.1371/journal.pone.0297284.g014

Graph: Fig 15 info:doi/10.1371/journal.pone.0297284.g015

Graph: Fig 16 info:doi/10.1371/journal.pone.0297284.g016

To address this challenge, this study introduces an improved segmentation strategy, specifically for high-level TPD images with scarce latex. This method initially performs multi-threshold segmentation using DBO-Otsu, followed by a morphological assessment of the tapping cut images to examine the segmentation outcome. If the segmented shape is not curvilinear, it is identified as a high-level TPD image. The strategy then utilizes the maximum non-zero value at the end of the grayscale histogram as the final latex segmentation threshold, thereby precisely locating the latex areas. This approach considers the high grayscale value but low pixel count characteristic of the latex, enhancing accuracy in identifying and segmenting sparse latex regions. Moreover, it effectively avoids missegmentation due to histogram noise or high grayscale values in non-latex areas, crucial for analyzing high-level TPD images as accurate extraction of latex areas is vital for disease assessment and subsequent processing. Re-segmenting the latex using this method, as represented in Figs 14–16, aligns the results with actual conditions, successfully extracting the latex images.

4 Experiments and analysis of results

4.1 Parameter settings of the algorithm

In order to assess the performance of the proposed DBO-Otsu algorithm, we randomly selected three images from the rubber dataset as benchmark images. Due to the stochastic nature of metaheuristic algorithms, the results vary with each execution. In this context, each algorithm was subjected to 50 experimental trials, and the results were then averaged. This method was compared for performance with the original Otsu method, SSA-Otsu method [[80]], WOA-Otsu method [[81]], WSO-Otsu method [[82]], GWO-Otsu method [[83]], AHA-Otsu method [[84]], and CSA-Otsu method [[85]]. The parameter settings for each algorithm are shown in Table 1. Except for pop_num and Max_iter, which were modified to accommodate the complexity of the experiments, the regional parameters were adopted from recommended studies.

Graph

Table 1 Parameter setting of the testing algorithm.

Algorithm	Parameter	Setting
SSA-Otsu	pop_num	60
	Max_iter	200
WOA-Otsu	pop_num	60
	Max_iter	200
WSO-Otsu	pop_num	60
	Max_iter	200
	fmax	0.75
	fmin	0.07
	tau	4.11
	pmin	0.5
	pmax	1.5
	a0	6.25
	a1	100
	a2	0.0005
GWO-Otsu	pop_num	60
	Max_iter	200
AHA-Otsu	pop_num	60
	Max_iter	50
CSA-Otsu	pop_num	60
	Max_iter	200
	rho	1
	p1	2
	p2	2
	c1	2
	c2	1.8
	gamma	2
	alpha	4
	beta	3
DBO-Otsu	pop_num	60
	Max_iter	200
	P_percent	0.2
	k	0.1
	b	0.3
	S	0.5

All these experiments were conducted using MATLAB R2022b on a Windows 11 operating system, with 16GB RAM memory and an Intel Core i5–11300 H CPU operating at 3.10 GHz.

4.2 Evaluation metrics

4.2.1 PSNR

Peak Signal-to-Noise Ratio (PSNR) is a widely used metric in image and video processing for objective quality assessment. It is defined as the ratio between the maximum possible power of a signal and the power of the noise that affects the fidelity of its representation. The PSNR is usually expressed in logarithmic decibel scale. The PSNR is calculated using the following formula:

Graph

$$\begin{array}{c}\hfill \text{PSNR}=10·{\text{log}}_{10}\left(\frac{MA{X}_I^2}{MSE}\right)\end{array}$$ (24)

Where: MAXI is the maximum possible pixel value of the image. MSE is the Mean Squared Error between the reference image and the distorted image.

4.2.2 FSIM

Feature Similarity Index (FSIM) is a more advanced metric that considers luminance, contrast, and structure similarities between the reference and distorted images to compute the similarity index. The FSIM is calculated using the following formula:

Graph

$$\begin{array}{c}\hfill \text{FSIM}=\frac{1}{N}\sum \frac{2·{\mu }_X·{\mu }_Y+C_1}{{\mu }_X^2+{\mu }_Y^2+C_1}·\frac{2·{\sigma }_{XY}+C_2}{{\sigma }_X^2+{\sigma }_Y^2+C_2}\end{array}$$ (25)

Where: μX and μY are the local means of images X and Y, respectively. σX and σY are the local standard deviations of images X and Y, respectively. σXY is the local cross covariance between images X and Y. C1 and C2 are constants. N is the total number of pixels in the images.

4.2.3 SSIM

Structural Similarity Index (SSIM) is another advanced metric for comparing the similarity between two images. The SSIM index is designed to improve on traditional metrics like PSNR and MSE by considering changes in structural information, luminance, and contrast. The SSIM is calculated using the following formula:

Graph

$$\begin{array}{c}\hfill \text{SSIM}(x,y)=\frac{(2·{\mu }_x·{\mu }_y+C_1)}{({\mu }_x^2+{\mu }_y^2+C_1)}·\frac{(2·{\sigma }_{xy}+C_2)}{({\sigma }_x^2+{\sigma }_y^2+C_2)}\end{array}$$ (26)

Where: μx and μy are the mean of images x and y, respectively. σx and σy are the variance of images x and y, respectively. σxy is the covariance of images x and y. C1 and C2 are constants used to avoid division by zero.

4.3 Indicator testing

In our comparative study, the DBO-Otsu algorithm was evaluated against six other advanced Otsu methodologies: SSA-Otsu, WOA-Otsu, WSO-Otsu, GWO-Otsu, AHA-Otsu, and CSA-Otsu. We employed several metrics for this assessment, including runtime, PSNR, FSIM, and SSIM. Runtime, as a measure of real-time performance, is a critical factor in gauging the efficiency of an algorithm. PSNR, FSIM, and SSIM, which are closely tied to the structural attributes of images, serve as indicators of the segmentation quality. Selected results from this comparative analysis are shown in Table 2, focusing on operational data for levels 4–6, 4–12, 4–19, and 4–20.

Graph

Table 2 Test result of seven algorithms.

Group	Algorithm	ScarThreshold	LatexThreshold	Time	PSNR	FSIM	SSIM
4–6	SSA	170.607846	202.3673229	0.454	8.149481674	0.514599926	0.126894101
	WOA	193.3105468	255.4191533	0.292	7.569149399	0.505307484	0.114336442
	WSO	174.129266	202.9681429	0.293	8.067329821	0.513251516	0.123459772
	GWO	170.9923696	202.4617205	0.297	8.149481674	0.514599926	0.126894101
	AHA	170.6434785	202.960859	0.326	8.149481674	0.514599926	0.126894101
	CSA	170.704867	202.3591813	0.293	8.149481674	0.514599926	0.126894101
	DBO	171.4228188	204.8375643	0.372	8.165899285	0.521412398	0.127876046
4–12	SSA	155.5854386	207.8677594	0.44	8.289015524	0.557163498	0.09489891
	WOA	149.0694241	203.5416501	0.293	15.17593885	0.590534328	0.550520428
	WSO	154.2989164	202.5409691	0.283	8.333545453	0.558486259	0.100172884
	GWO	157.7155776	205.8915754	0.3	8.224993212	0.556169337	0.093358665
	AHA	155.2227337	207.6226922	0.332	8.289015524	0.557163498	0.09489891
	CSA	155.304461	207.5493689	0.297	8.289015524	0.557163498	0.09489891
	DBO	152.3887825	204.0333982	0.335	8.364000908	0.557133382	0.100182992
4–19	SSA	171.4507	193.7683	0.45	11.44445	0.444172	0.074768
	WOA	171.3533	193.0518	0.357	11.37516	0.432095	0.071192
	WSO	166.0914	195.5994	0.338	11.49231	0.440329	0.076701
	GWO	171.3085	193.8558	0.931	11.44445	0.444172	0.074768
	AHA	171.8939	193.1976	0.596	11.44445	0.444172	0.074768
	CSA	168.8158	193.9061	0.366	11.51899	0.452194	0.078359
	DBO	166.6904	193.8078	0.462	11.54439	0.452779	0.079371
4–20	SSA	197.4308	209.5352	0.601	6.124981	0.60167	0.041313
	WOA	182.7351	208.6473	0.4	11.72238	0.613376	0.562652
	WSO	193.5352	213.5405	0.871	6.128787	0.606108	0.045566
	GWO	197.0394	209.767	0.491	6.124981	0.60167	0.041313
	AHA	197.4107	209.7839	0.405	6.124981	0.60167	0.041313
	CSA	197.0325	209.6747	0.361	6.124981	0.60167	0.041313
	DBO	181.2085	208.6152	0.464	7.256473	0.641461	0.133997

1 Note: Bold indicates the best score for each item

In terms of runtime, the DBO-Otsu algorithm demonstrated superior performance over other enhanced Otsu methods, maintaining moderate processing times in all experimental setups. In the analysis of the DBO-Otsu method, a comprehensive evaluation was conducted, focusing on the average PSNR, FSIM, and SSIM scores under diverse experimental conditions.

The assessment in the Table 3 revealed that the DBO-Otsu method consistently demonstrated high performance. Specifically, it achieved the highest average ranking in both PSNR and SSIM scores, with an impressive average rank of 1.50 for each. In the FSIM category, DBO-Otsu also performed commendably, securing an average rank of 2.25. These rankings underscore its proficiency in several key areas, particularly in one of the experimental domains where it excelled. The integration of the DBO-Otsu method with the advanced DBO algorithm has been instrumental in enhancing segmentation accuracy.

Graph

Table 3 Final average rankings of algorithms in PSNR, FSIM, and SSIM.

Algorithm	Average PSNR Ranking	Average FSIM Ranking	Average SSIM Ranking
SSA	4.75	4.25	4.75
WOA	4.00	4.25	4.00
WSO	3.75	4.25	3.75
GWO	5.25	5.00	5.25
AHA	4.75	4.25	4.75
CSA	4.00	3.75	4.00
DBO	1.50	2.25	1.50

The statistical significance of DBO-Otsu's performance was determined using a Wilcoxon signed-rank test at a significance level of 0.1. The outcomes in Table 4 revealed statistically significant results for the DBO-Otsu method in terms of PSNR and SSIM, whereas the FSIM scores did not reach a similar level of statistical significance. These results confirm the effectiveness of the DBO-Otsu method and highlight areas for potential refinement.

Graph

Table 4 Wilcoxon test results comparing DBO with other algorithms.

Algorithm	PSNR p-value	FSIM p-value	SSIM p-value
SSA	0.0625	0.1250	0.0625
WOA	0.8125	0.4375	0.8125
WSO	0.0625	0.1250	0.0625
GWO	0.0625	0.0625	0.0625
AHA	0.0625	0.1250	0.0625
CSA	0.0625	0.1250	0.0625

To visually illustrate the efficacy of various optimization algorithms in threshold optimization, we generated convergence curves, as depicted in Fig 17. The horizontal axis on these curves represents the number of iterations, while the vertical axis reflects the best fitness value achieved to date.

Graph: Fig 17 info:doi/10.1371/journal.pone.0297284.g017

As can be seen in Fig 17, although other algorithms like SSA demonstrated superior final results in some experiments (such as 4–8, 4–19, and 4–20), DBO-Otsu exhibited strong performance in both convergence speed and accuracy, which was particularly evident in most of the tested functions.

4.4 Detail verification

In Table 1, we present the selected thresholds and describe the division of the image into five regions based on these thresholds, with each pixel's value determined by its corresponding region. The detailed assessment process includes: firstly applying multi-threshold processing on the original image using different methods, then selecting regions containing key information for comparison. Lastly, we calculate the difference in grayscale values for each pixel in these regions compared to the corresponding areas in the original image to determine a detail score.

To more clearly demonstrate and evaluate these details, specific regions were analyzed in Figs 18–20, which contain detailed information about the rubber tree tapping cuts (see Fig 21).

Graph: Fig 18 info:doi/10.1371/journal.pone.0297284.g018

Graph: Fig 19 info:doi/10.1371/journal.pone.0297284.g019

Graph: Fig 20 info:doi/10.1371/journal.pone.0297284.g020

Graph: Fig 21 info:doi/10.1371/journal.pone.0297284.g021

Furthermore, we established a scoring system for curve similarity as follows:

Graph

$$\begin{array}{c}\hfill \begin{array}{cc}\hfill \text{Score}& ={\displaystyle \sum _{i=1}^L}\text{score}\left(i\right)\hfill \\ \hfill score\left(i\right)& =\{\begin{array}{c}1,\left|gray\right(i)-gray1(i\left)\right|\le 50\hfill \\ -1,\left|gray\right(i)-gray1(i\left)\right|>50\hfill \end{array}\hfill \end{array}\end{array}$$ (27)

Where i is the index of the pixel in the selected region, L is the total number of pixels in that region, gray1(i) represents the grayscale value of the processed curve at the ith point, and gray(i) is the grayscale value of the original curve at the ith point.

The score increases when the difference in values at a given point between the two curves is small; it decreases when the difference is large. The total scores are then summed to obtain a final score. By comparing the scores in Table 3, we detailed the segmentation results of various algorithms.

The score increases when the difference in values at a given point between the two curves is small; it decreases when the difference is large. The total scores are then summed to obtain a final score. By comparing the scores in Table 5, we detailed the segmentation results of various algorithms.

Graph

Table 5 Fitness score table.

Algorithm	A 3–8	B 3–7	C 1–3	Mean
SSA-Otsu	-816	960	186	110
Otsu	-414	1944	846	792
WOA-Otsu	-816	972	234	130
WSO-Otsu	-600	522	468	130
GWO-Otsu	-822	1002	474	218
AHA-Otsu	-840	1002	60	74
CSA-Otsu	-750	858	102	70
DBO-Otsu	636	1248	2874	1586

In each set of experiments, DBO-Otsu scored the highest in detail retention, demonstrating its superiority in preserving original image details compared to other algorithms.

4.5 Application evaluation

To ascertain the DBO-Otsu algorithm's practical effectiveness developed in this research, we conducted segmentation tests using images from each level of the dataset, as illustrated in Fig 22.

Graph: Fig 22 info:doi/10.1371/journal.pone.0297284.g022

The segmentation of tapping cuts, excluding those in level 3 images, proved effective in the remaining images, as evidenced in Fig 23. The level 3 images, characterized by blurred feature boundaries, presented a challenge, where the automated segmentation approach may not have been entirely suitable, resulting in less than optimal outcomes.

Graph: Fig 23 info:doi/10.1371/journal.pone.0297284.g023

Furthermore, latex segmentation was executed on the original images, yielding the results shown in Fig 24. These results demonstrate that the DBO-Otsu algorithm successfully segments images even at levels 4 and 5, where latex pixels are sparse, thus overcoming the traditional Otsu method's limitations in handling areas with scant grayscale pixels. In the case of level 1, 2, and 3 images, some missegmentation occurred. However, these inaccuracies were addressed in the final statistical analysis, ensuring the overall results remained within an acceptable margin.

Graph: Fig 24 info:doi/10.1371/journal.pone.0297284.g024

As defined in the criteria for calculating TPD levels in the reference [[86]], the ratio of latex to tapping cuts is designated as λ = S1/S2, where S1 represents the area of latex and S2 denotes the area of tapping cuts. Post-segmentation, the black areas in the images represent the background, while the white areas signify the regions of interest. The ratio of λ can be approximated by calculating the proportion of pixels covering both the tapping cuts and the latex. The criteria for classifying different levels of TPD in this study are detailed in Table 6.

Graph

Table 6 TPD grade determination table.

Latex/Incision Ratio (λ)	Determination Grade
1% ≤λ ≤ 10%	Grade 5
10% <λ ≤ 20%	Grade 4
20% <λ ≤ 35%	Grade 3
35% <λ ≤ 45%	Grade 2
λ> 45%	Grade 1

Utilizing MATLAB software, we quantified the number of pixels pertaining to tapping cuts and latex in Figs 23 and 24. This analysis enabled the determination of TPD levels based on their ratio, as depicted in Table 7.

Graph

Table 7 Level determination table.

levels	Image Number	Latex Pixel Count	Scar Pixel Count	Area Ratio	Disease Level
level 1	13	7681	11099	0.692044328	1
	14	13842	24797	0.558212687	1
	15	11545	18598	0.620765674	1
level 2	4	41651	96132	0.433268839	2
	10	182369	492541	0.370261562	2
	26	263777	612148	0.430903964	2
level 3	10	175705	547090	0.32116288	3
	11	173405	590372	0.293721586	3
	12	147825	505293	0.292553034	3
level 4	11	15753	194899	0.080826479	5
	12	60878	525198	0.115914379	4
	16	19962	193855	0.102973872	4
level 5	16	126	12723	0.009903325	5
	17	128	11821	0.010828187	5
	18	189	399566	0.000473013	5

The algorithm demonstrated exceptional proficiency in segmenting both latex and tapping cuts. It precisely identified the TPD levels through the calculated ratios, offering accuracy and efficiency that surpass traditional manual identification methods, thereby significantly reducing labor costs.

Additionally, we conducted a random evaluation of images across different TPD levels within the dataset, analyzing 15–20 images per level. The findings of this evaluation are summarized in the Table 8:

Graph

Table 8 Accuracy rate table.

Judgment Level	Image Count	Accuracy Rate	Average Accuracy Rate
1	26	92%	84.43%
2	25	92%
3	20	65%
4	20	81%
5	20	88%

These results reveal that the segmentation performance of this method in level 1–2 and 4–5 images is markedly superior to that in level 3 images, with an accuracy rate exceeding 80%. The diminished accuracy observed in level 3 images is attributed to the nuances in segmentation accuracy. Higher-level image segmentation accuracy declines with an increasing area ratio, whereas for lower-level images, it decreases with a decreasing area ratio. This results in a moderate latex-to-tapping cut ratio for level 3 images, posing a challenge for accurate segmentation identification.

5 Conclusions

The primary aim of this study was to enhance image segmentation by refining the classic Otsu thresholding method, with a specific focus on preserving intricate details. The motivation behind this research was to address the challenge of exponential time complexity growth in multi-level threshold computations. To achieve this, we introduced the innovative DBO algorithm, which was integrated into the Otsu method to create the DBO-Otsu algorithm—a novel image segmentation tool.

Our rigorous performance evaluation of the DBO-Otsu algorithm encompassed a comprehensive set of performance metrics, including PSNR, FSIM, and SSIM. The results demonstrated that DBO-Otsu not only maintained computational efficiency but also significantly reduced image distortion. In fact, DBO-Otsu surpassed the performance of six other comparative methods in preserving image structural integrity.

In practical applications of image segmentation, we encountered variations in latex quantities across different rubber tree disease levels. It became evident that a direct application of DBO-Otsu might not suffice for all scenarios. Therefore, we adopted a nuanced approach by conducting morphological analyses post-initial segmentation and adapting strategies tailored to images at various disease stages. While accuracy experienced a slight decline in images of intermediate disease levels, the majority of judgments remained acceptably accurate, with minimal errors.

In conclusion, our findings underscore the importance of prioritizing the DBO-Otsu algorithm in future research endeavors, especially in contexts where rapid and efficient TPD diagnosis is paramount. Notably, in instances with pronounced disease symptoms, the DBO-Otsu algorithm has the potential to deliver even more remarkable results. This approach not only expedites computation but also upholds high image quality, presenting a robust and efficient solution in image segmentation. However, it is crucial to acknowledge the inherent limitations within the algorithm, which excel in diagnosis based on relative proportions but may still face challenges in isolating specific targets. Our future research will be dedicated to enhancing segmentation accuracy, including the potential incorporation of edge detection algorithms to eliminate irrelevant areas on the trunk.

Supporting information

S1 Data

(ZIP)

S2 Data

(ZIP)

S3 Data

(ZIP)

S4 Data

(ZIP)

S5 Data

(ZIP)

S6 Data

(ZIP)

S7 Data

(ZIP)

S8 Data

(ZIP)

S9 Data

(ZIP)

S10 Data

(ZIP)

S1 File

(ZIP)

S2 File

(ZIP)

Decision Letter 0

Khan Khan Bahadar Academic Editor

2 Nov 2023

PONE-D-23-30775Advancing Image Segmentation with DBO-Otsu: Addressing Rubber Tree Diseases through Enhanced Threshold TechniquesPLOS ONE

Dear Dr. Liu,

Thank you for submitting your manuscript to PLOS ONE. After careful consideration, we feel that it has merit but does not fully meet PLOS ONE's publication criteria as it currently stands. Therefore, we invite you to submit a revised version of the manuscript that addresses the points raised during the review process.

Please submit your revised manuscript by Dec 17 2023 11:59PM. If you will need more time than this to complete your revisions, please reply to this message or contact the journal office at plosone@plos.org. When you're ready to submit your revision, log on to https://www.editorialmanager.com/pone/ and select the 'Submissions Needing Revision' folder to locate your manuscript file.

Please include the following items when submitting your revised manuscript:

If you would like to make changes to your financial disclosure, please include your updated statement in your cover letter. Guidelines for resubmitting your figure files are available below the reviewer comments at the end of this letter.

If applicable, we recommend that you deposit your laboratory protocols in protocols.io to enhance the reproducibility of your results. Protocols.io assigns your protocol its own identifier (DOI) so that it can be cited independently in the future. For instructions see: https://journals.plos.org/plosone/s/submission-guidelines#loc-laboratory-protocols. Additionally, PLOS ONE offers an option for publishing peer-reviewed Lab Protocol articles, which describe protocols hosted on protocols.io. Read more information on sharing protocols at https://plos.org/protocols?utm_medium=editorial-email&utm_source=authorletters&utm_campaign=protocols.

We look forward to receiving your revised manuscript.

Kind regards,

Khan Bahadar Khan, Ph.D

Academic Editor

PLOS ONE

Journal Requirements:

When submitting your revision, we need you to address these additional requirements.

1. Please ensure that your manuscript meets PLOS ONE's style requirements, including those for file naming. The PLOS ONE style templates can be found at 

https://journals.plos.org/plosone/s/file?id=wjVg/PLOSOne%5fformatting%5fsample%5fmain%5fbody.pdf and 

https://journals.plos.org/plosone/s/file?id=ba62/PLOSOne_formatting_sample_title_authors_affiliations.pdf

• 2. Please note that PLOS ONE has specific guidelines on code sharing for submissions in which author-generated code underpins the findings in the manuscript. In these cases, all author-generated code must be made available without restrictions upon publication of the work. Please review our guidelines at https://journals.plos.org/plosone/s/materials-and-software-sharing#loc-sharing-code and ensure that your code is shared in a way that follows best practice and facilitates reproducibility and reuse.
• 3. Thank you for stating the following financial disclosure: 

"Hainan Provincial Natural Science Foundation of China (623RC449)"  

Please state what role the funders took in the study.  If the funders had no role, please state: ""The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript."" 

If this statement is not correct you must amend it as needed. 

Please include this amended Role of Funder statement in your cover letter; we will change the online submission form on your behalf.

4. In your Data Availability statement, you have not specified where the minimal data set underlying the results described in your manuscript can be found. PLOS defines a study's minimal data set as the underlying data used to reach the conclusions drawn in the manuscript and any additional data required to replicate the reported study findings in their entirety. All PLOS journals require that the minimal data set be made fully available. For more information about our data policy, please see http://journals.plos.org/plosone/s/data-availability.

""Upon re-submitting your revised manuscript, please upload your study's minimal underlying data set as either Supporting Information files or to a stable, public repository and include the relevant URLs, DOIs, or accession numbers within your revised cover letter. For a list of acceptable repositories, please see http://journals.plos.org/plosone/s/data-availability#loc-recommended-repositories. Any potentially identifying patient information must be fully anonymized.

Important: If there are ethical or legal restrictions to sharing your data publicly, please explain these restrictions in detail. Please see our guidelines for more information on what we consider unacceptable restrictions to publicly sharing data: http://journals.plos.org/plosone/s/data-availability#loc-unacceptable-data-access-restrictions. Note that it is not acceptable for the authors to be the sole named individuals responsible for ensuring data access.

We will update your Data Availability statement to reflect the information you provide in your cover letter.

• 5. We note that you have stated that you will provide repository information for your data at acceptance. Should your manuscript be accepted for publication, we will hold it until you provide the relevant accession numbers or DOIs necessary to access your data. If you wish to make changes to your Data Availability statement, please describe these changes in your cover letter and we will update your Data Availability statement to reflect the information you provide.
• 6. Please ensure that you refer to Figure 1, 2, 12, 13 and 14 in your text as, if accepted, production will need this reference to link the reader to the figure.
• 7. Please include captions for your Supporting Information files at the end of your manuscript, and update any in-text citations to match accordingly. Please see our Supporting Information guidelines for more information: http://journals.plos.org/plosone/s/supporting-information.

[Note: HTML markup is below. Please do not edit.]

Reviewers' comments:

Reviewer's Responses to Questions

Comments to the Author

1. Is the manuscript technically sound, and do the data support the conclusions?

The manuscript must describe a technically sound piece of scientific research with data that supports the conclusions. Experiments must have been conducted rigorously, with appropriate controls, replication, and sample sizes. The conclusions must be drawn appropriately based on the data presented.

Reviewer #1: Yes

Reviewer #2: Partly

***

2. Has the statistical analysis been performed appropriately and rigorously?

Reviewer #1: Yes

Reviewer #2: No

***

3. Have the authors made all data underlying the findings in their manuscript fully available?

The http://www.plosone.org/static/policies.action#sharing requires authors to make all data underlying the findings described in their manuscript fully available without restriction, with rare exception (please refer to the Data Availability Statement in the manuscript PDF file). The data should be provided as part of the manuscript or its supporting information, or deposited to a public repository. For example, in addition to summary statistics, the data points behind means, medians and variance measures should be available. If there are restrictions on publicly sharing data—e.g. participant privacy or use of data from a third party—those must be specified.

Reviewer #1: Yes

Reviewer #2: No

***

4. Is the manuscript presented in an intelligible fashion and written in standard English?

PLOS ONE does not copyedit accepted manuscripts, so the language in submitted articles must be clear, correct, and unambiguous. Any typographical or grammatical errors should be corrected at revision, so please note any specific errors here.

Reviewer #1: Yes

Reviewer #2: Yes

***

5. Review Comments to the Author

Please use the space provided to explain your answers to the questions above. You may also include additional comments for the author, including concerns about dual publication, research ethics, or publication ethics. (Please upload your review as an attachment if it exceeds 20,000 characters)

Reviewer #1: In this paper, combining the DBO algorithm with the Otsu method, a new DSPO-OTSU image segmentation algorithm for rubber tree tangent and latex segmentation is proposed. In a comparative evaluation with six other meta-heuristic OTSU algorithms, namely SSA-OTSU, WOA-OTSU, WSO-OTSU, GGO-OTSU, AHA-OTSU, and CSA-OTSU, the results showed a 10% improvement in the structural similarity index.

Section 1.2 presents some challenges to this approach, but these challenges are not properly discussed in the results. In this paper, the DBO-Otsu method for rubber tree tangential and latex segmentation is introduced only by comparing the tangential results without giving a detailed analysis of latex segmentation. The paper also has the following problems：

m(k) in formula (0.6) represents m1(k) or m2(k)? Or some other meaning? Please give a specific explanation;

The specific meaning of the mathematical symbols in formula (8) , e.g., Jtmax, sigmoid and w need to be clarified.

Table 1 in section 4.4 is not found;

The introduction to Figure 1 is not presented;

The segmentation diagram of the traditional Otsu algorithm is shown in Figure 5, but the segmentation diagram of the DPO-OTSU method proposed in this paper is not given.

Figure 6 shows three different convergence graphs, but does not explain the difference between the three convergence graphs, only compares the difference between the DBO algorithm and other algorithms;

Figure 10-12 corresponds to the segmentation results of Figure 7-9 by different algorithms, but it is not introduced which original figure of Figure 7-9 corresponds to each column of results.

Does Figure 12 in line 9 on page 26 correspond to Figure 2 on page 36? Is it marked incorrectly?

Does row 7 in Figure 13 from the bottom of page 26 correspond to Figure 3 on page 37? Is it marked incorrectly?

The delivered figure is too blurred with lower resolution to distinguish. Please provide clearer figures and diagrams for readers. Meanwhile, a paper related to the rubber tree disease could be mentioned in the Introduction Section, which is mentioned as follows: Rubber Tree Crown Segmentation and Property Retrieval Using Ground-Based Mobile LiDAR after Natural Disturbances.

Reviewer #2: This paper presents a multi-level thresholding image segmentation model, DBO-Otsu, designed to address the need for precise classification and early diagnosis of Rubber Tree Top Wilt Disease. DBO-Otsu, based on Dung Beetle Optimization, outperforms standard Otsu methods, especially in high grayscale areas. The proposed DBO-Otsu is evaluated over three images from the rubber dataset as benchmark images. Comparative analysis with six other meta-heuristic Otsu algorithms shows a notable 10% enhancement in the Structural Similarity Index.

The paper highlights the utilization of image segmentation methods as a preprocessing step in the field of medical imaging, focusing on a timely and critical issue related to healthcare diagnostics.

The reviewer appreciates the effort invested in this manuscript. While the paper exhibits notable merits, there are significant improvements that should be considered before the paper becomes suitable for publication. Therefore, the reviewer recommends addressing these additional revisions.

Please see the attached file for detailed comments.

***

6. PLOS authors have the option to publish the peer review history of their article (https://journals.plos.org/plosone/s/editorial-and-peer-review-process#loc-peer-review-history). If published, this will include your full peer review and any attached files.

If you choose "no", your identity will remain anonymous but your review may still be made public.

Do you want your identity to be public for this peer review? For information about this choice, including consent withdrawal, please see our https://www.plos.org/privacy-policy.

Reviewer #1: No

Reviewer #2: No

***

[NOTE: If reviewer comments were submitted as an attachment file, they will be attached to this email and accessible via the submission site. Please log into your account, locate the manuscript record, and check for the action link "View Attachments". If this link does not appear, there are no attachment files.]

While revising your submission, please upload your figure files to the Preflight Analysis and Conversion Engine (PACE) digital diagnostic tool, https://pacev2.apexcovantage.com/. PACE helps ensure that figures meet PLOS requirements. To use PACE, you must first register as a user. Registration is free. Then, login and navigate to the UPLOAD tab, where you will find detailed instructions on how to use the tool. If you encounter any issues or have any questions when using PACE, please email PLOS at figures@plos.org. Please note that Supporting Information files do not need this step.

Attachment

Submitted filename: review_report_PONE-D-23-30775.pdf

Author response to Decision Letter 0

29 Nov 2023

Zhenjing Xie

Hainan University

Haikou, 570228

China

2023.11.28

Khan Bahadar Khan, Ph.D

Academic Editor

PLOS ONE

Dear Editor Khan Bahadar Khan,

Thank you and the reviewers for the review and recommendations on our manuscript (Title: "Advancing Image Segmentation with DBO-Otsu: Addressing Rubber Tree Diseases through Enhanced Threshold Techniques," Manuscript ID: PONE-D-23-30775). We have made detailed revisions based on the reviewer's feedback. Below is our response to the reviewer's comments:

Major comments:

Introduction:

Comment: Begin the introduction with a brief overview of the Rubber Tree Top Wilt Disease and how cutting-edge AI techniques and computer vision play a pivotal role in identifying and understanding this disease. This will allow you to smoothly introduce image segmentation as a preprocessing step in medical imaging, setting the foundation for the subsequent discussion.

Response: Thank you for your detailed review and valuable suggestions regarding the introduction section of our paper. We recognize that the original introduction lacked the necessary cohesion, flow, and clarity of ideas. Following your guidance, we have thoroughly revised the introduction to better align with the theme and purpose of the paper.

In the revised introduction, we start with a brief overview of Rubber Tree Topping Panel Dryness (TPD), highlighting its significance to the global rubber industry. We further discuss the critical role of modern AI techniques and computer vision in identifying and understanding TPD. This approach allows us to smoothly introduce image segmentation as a preprocessing step in medical imaging, setting the foundation for the subsequent discussion.

Moreover, we emphasize the fundamental goal in image recognition, especially in dealing with the unique image characteristics of the cuts and latex, which is to achieve precise segmentation. High-quality image segmentation directly contributes to improved diagnostic accuracy, fostering objectivity and uniformity in evaluations. This approach aids in the automatic identification of TPD-affected rubber trees at different stages, assisting researchers in real-time monitoring and future latex yield prediction, and advancing the study of rubber tree diseases.

2. Comment: The authors mentioned that (traditional Otsu algorithms often face challenges in complex scenarios) Please clearly highlight the limitations of traditional Otsu algorithms in complex scenarios, to underscore the motivation for adopting metaheuristics as an alternative approach.

Response: Thank you very much for your thorough review and suggestions regarding our mention of the limitations of traditional Otsu algorithms in the introduction section of our paper. We realize that our original text did not sufficiently elucidate the limitations of traditional Otsu algorithms in complex scenarios. Therefore, following your guidance, we have enhanced and clarified this part in our revised manuscript.

In the revised introduction, we discuss in more detail the performance limitations of the Otsu algorithm in scenarios with highly variable background and target intensity, or significant noise disturbances. Particularly, its robustness is compromised in high-noise images affected by salt-and-pepper noise, leading to segmentation inaccuracies. We highlighted the novel improvements proposed to address these challenges.

We hope these revisions clearly underscore the motivation for adopting metaheuristic algorithms as an alternative approach and better explain the limitations of traditional Otsu algorithms.

3. Comment: Introduce thresholding-based techniques, emphasizing their popularity and relevance as similarity-based methods in the context of your research.

Response: Thank you very much for your suggestion regarding the introduction of thresholding-based techniques and emphasizing their popularity and relevance as similarity-based methods in the context of our research. We acknowledge that our original text did not adequately highlight this aspect. Therefore, following your guidance, we have made relevant enhancements and clarifications in our revised manuscript.

In the revised introduction, we have elaborated on threshold-based image segmentation techniques, emphasizing their simplicity and effectiveness, and their applicational value in our current research. Specifically, we mention that despite the superior performance of deep learning methods in image segmentation, their hardware requirements limit their application in wearable devices. Therefore, for Tapping Panel Dryness (TPD) recognition in rubber plantations, where environmental complexity and signal instability are issues, the lighter, more efficient threshold-based segmentation methods become a more appropriate choice.

We believe these additions and improvements not only provide a comprehensive presentation of the importance of threshold-based image segmentation techniques but also better contextualize them in our research.

4. Comment: Provide a more detailed explanation for the selection of Dung Beetle Optimization as the chosen metaheuristic approach and discuss why it was preferred over other available metaheuristics.

Response: Thank you very much for pointing out the need for a more detailed explanation of why we chose Dung Beetle Optimization (DBO) as the metaheuristic approach and its advantages over other available metaheuristic algorithms. We realize that our original text did not adequately articulate the specific reasons for this choice. Therefore, following your suggestion, we have made corresponding supplements and detailed explanations in our revised manuscript.

In the revised introduction, we elaborate on why we chose to combine the DBO algorithm with the Otsu method and its advantages in handling complex image segmentation challenges, such as rubber tree cuts and latex images. We emphasize that the DBO algorithm is inspired by the natural behaviors of dung beetles, such as their rolling and foraging strategies, which enhance the algorithm's performance in terms of search diversity, efficiency, and global convergence capabilities. Especially in multi-threshold image segmentation, DBO excels in convergence speed and solution accuracy compared to other metaheuristic algorithms.

We hope these additions clearly articulate why we chose the DBO algorithm as our optimization tool and better contextualize it in our research.

5. Comment: It is difficult to discern the unique contributions of the proposed model. It would be beneficial to highlight the contributions in bullet points. This would provide readers with a concise and structured overview of the specific advancements by the proposed model.

Response: Thank you for pointing out that the unique contributions of our proposed model were not clearly discernible and for suggesting that we highlight these contributions in bullet points. Indeed, our original text did not clearly list the specific advancements of the proposed model. Therefore, following your advice, we have highlighted these contributions in a more structured and concise manner in our revised manuscript.

In the revised introduction, we clearly outline the paper's four primary contributions in a bulleted list: (1) Introducing the DBO-Otsu algorithm to enhance the efficiency and accuracy of complex image segmentation; (2) Conducting a comprehensive evaluation of the DBO-Otsu algorithm using mainstream performance metrics; (3) Providing a theoretical analysis of the DBO algorithm, emphasizing its promising applications in image segmentation; (4) Demonstrating the algorithm's practical utility in real-world segmentation tasks, particularly in diagnosing rubber tree Tapping Panel Dryness. We also outline the structure of the paper.

In addition to the previously mentioned four primary contributions, we have further enriched the content of our paper. To more vividly demonstrate the advantages of our algorithm, the experimental steps, and the core strategies, we have included a comprehensive flowchart. This flowchart not only clearly delineates the operational process of the DBO-Otsu algorithm but also visually presents the various stages of the experiment and the detailed implementation of key strategies.

With this flowchart, we aim to provide readers with a more intuitive understanding, allowing them to grasp the overall picture and advantages of our method at a glance. We believe that this graphical representation not only enhances the readability of the paper but also helps deepen readers' understanding of our research.

Inadequate literature review

1. Comment: discussion of related works should be expanded with more recent metaheuristics-based image segmentation studies to highlight the limitations of the existing literature, situate the contribution, establish the research gap and highlight the novelty of the proposed approach.

Response: Thank you for suggesting that we expand the discussion of the integration of metaheuristic algorithms with the Otsu method in image segmentation to highlight the limitations of the existing literature, establish the research gap, and emphasize the novelty of our proposed approach. We realize that our original text did not sufficiently showcase these latest studies. Therefore, following your advice, we have included more recent related research in section 1.3 of our revised manuscript.

In these additions, we have detailed how the combination of metaheuristic algorithms with the Otsu method significantly enhances its capabilities. We listed a series of innovative studies such as DE-OTSU-GWO, FOA-OTSU, IGJO, AOA optimization of the Otsu method, LCGSA, and HCROA, which have made significant achievements in improving the accuracy, efficiency, and stability of image segmentation. We also emphasized how the DBO-Otsu algorithm, inspired by the natural behaviors of dung beetles, enhances search diversity, efficiency, and global convergence capabilities, outperforming other metaheuristic algorithms in multi-threshold image segmentation, particularly in convergence speed and solution accuracy.

We hope these additions provide readers with a comprehensive perspective on the current developments in the research field and clearly highlight the contributions and innovation of our proposed method.

2. Comment: Some recent related papers that could be included.

Response: Thank you for suggesting that we include recent related research papers in the introduction section of our manuscript. We realize that the inclusion of these references will enrich the research background of our paper and better showcase the current advancements in the field. Following your guidance, we have added several key recent papers in section 1.3 of our revised manuscript.

In these additions, we particularly highlight the latest advancements in swarm intelligence algorithms for multi-threshold segmentation, especially in their application to processing COVID-19 chest X-rays and CT scans. We have cited studies by Chen et al., Abualigah et al., Liu et al., Emam et al., and Chen et al., who have made significant contributions in enhancing initial convergence, global search efficiency, search efficiency, and segmentation precision, as well as in multi-threshold segmentation applications. These advancements not only advance the application of swarm intelligence in medical image processing but also offer potent tools for medical decision-making.

We hope these additions provide readers with a comprehensive perspective on the current developments in the research field. Thank you once again for your valuable suggestions, which have been instrumental in enhancing the depth and breadth of our work.

• 3. DBO-Otsu

1. Comment: Please expand the DBO-Otsu approach in Section 3, explaining how DBO is integrated with Otsu. Clarify the solution representation and fitness function. Consider using schematic views and visualizations for better clarity.

Response: We have expanded upon the DBO-Otsu method in Section 3, explaining how DBO is integrated with Otsu, and providing clear representations of the solution and the fitness function. We have used flowcharts for visualization to facilitate better elucidation.

2. Comment: Section 3.1 is a critical component of the manuscript which represent the main contribution. Hence, it is essential to highlight and justify the proposed enhancements.

Response: In Section 3.1, we have highlighted and substantiated the significance of the algorithmic enhancements proposed, where the improved formula is repeatedly applied in the subsequent pseudocode. The efficacy of these improvements will be further evidenced through experimental comparisons with the traditional OTSU algorithm, accentuating their importance. Moreover, in Section 3.2, we delve into the detailed integration principles of the DBO algorithm with the OTSU algorithm. Additionally, in Section 3.3, we discuss the judgment methods for specific levels and the refined strategies for segmenting latex, complemented by illustrative images to demonstrate their effective outcomes.

• 4. Experiments and analysis of results

1. Comment: It is not clear from the manuscript whether the parameter settings have been tuned specifically for your research or if they are adopted from recommended studies.

Response:

In response to the issues you pointed out, we have made necessary revisions to enhance the clarity and technical depth of the paper. Specifically, we have provided more detailed explanations of the parameter settings used in the experiments.

In the original manuscript, we indeed did not explicitly indicate whether the parameter settings were specifically tailored for our research or adopted from recommended studies. To address this, we have explicitly stated in the revised manuscript that, apart from 'pop_num' (population size) and 'Max_iter' (maximum number of iterations), the other regional parameters were selected based on recommended research. We have adjusted 'pop_num' and 'Max_iter' to accommodate the complexity of the experiments while ensuring the reliability of our study under these specific conditions.

2. Comment: It is important to cite the utilized algorithms, specifying which versions of SSA, CSA, WOA, GWO, WSO, and AHA have been used in this research?

Response: To provide a more comprehensive comparative perspective, we have added a comparison with the CSA-Otsu method in the paper. All algorithms used for comparison, including SSA-Otsu, WOA-Otsu, WSO-Otsu, GWO-Otsu, AHA-Otsu, and CSA-Otsu, are accompanied by corresponding references, allowing readers to refer to the original research for the source and rationale of these parameter settings.

3. Comment: Please provide a brief description of the rubber dataset used as a benchmark in this study.

Response: We have provided a brief description of the rubber dataset used as the baseline, including its levels of Tapping Panel Dryness (TPD), and whether it was randomly selected or chosen under certain conditions.

4. Comment: While the selection of three images from the rubber dataset is a good starting point, it's important to note that this sample size may not be sufficient to comprehensively verify the efficiency of the proposed model. It would be beneficial, if possible, to expand the testing to include a more substantial number of images to ensure robust evaluation

Response: We have expanded the number of test samples to ensure comprehensive validation of the proposed model. This includes validation for both latex segmentation and tapping line segmentation. Furthermore, we have performed latex and tapping line segmentation on all images in the dataset, selected representative images for presentation in the paper, and provided the remaining images in the supplementary materials.

5. Comment: Please discuss the convergence curves of the proposed approach.

Response: We appreciate your valuable suggestions. Regarding the discussion of the convergence curves of the proposed DBO-Otsu method, we have provided a more detailed description and analysis in the revised manuscript.

In the original text before revision, we described the overall trend of the convergence curves and highlighted the rapid convergence speed of the DBO algorithm. However, upon your recommendation, we realized that the original description might not adequately showcase the detailed comparisons between various algorithms and the specific performance of the DBO-Otsu algorithm.

Therefore, in the revised manuscript, we not only present the convergence curves of different optimization algorithms but also provide in-depth analysis of these curves. Specifically, in Fig17, we explicitly demonstrate the convergence process of different algorithms and extensively discuss the performance of DBO-Otsu relative to other algorithms, such as SSA, under different experimental conditions. We found that, while in some experiments (e.g., 4-8, 4-19, and 4-20), other algorithms like SSA may perform better in the final results, DBO-Otsu exhibits strong performance in terms of convergence speed and accuracy, especially in the majority of test functions.

With these modifications, we aim to comprehensively showcase the performance of the DBO-Otsu algorithm and provide deeper insights into its performance under different conditions. We believe that these additions will help readers better understand the effectiveness and practicality of our approach.

6. Comment: Inadequate analysis and discussion of results: The results are not thoroughly analyzed. A more in-depth analysis of the results, including statistical analysis is needed.

Response: We conducted a more in-depth analysis of the results and applied the proposed method in practical scenarios. By statistically calculating the ratio of latex to tapping line pixels in the segmented images, we approximated the basis for assessing the level of Tapping Panel Dryness (TPD). We performed diagnosis on all the images in the dataset, and the diagnostic results are saved in Excel files within the folders corresponding to each level of TPD.

7. Comment: Statistical analysis is important to judge the significance of the findings.

Response: We have maximized our efforts to perform statistical analysis of the data for each experimental step and presented it in tabular form in the paper, with the excess data stored in the supplementary materials. In addition to increasing the statistical analysis of the previous experiments, we also referred to statistical analysis methods from other papers for use in the 'Application Evaluation' section, aiming to achieve clearer experimental results.

Conclusion:

Comment: The conclusion should be explored better and it needs to contemplate the eventual restrictions of the developed technique to address future works in this area.

Response: Thank you for your insightful comments and valuable suggestions regarding the conclusion section of our paper. We realize that the original manuscript did not adequately explore the potential limitations of the DBO-Otsu algorithm, nor did it thoroughly consider future directions in this field. Therefore, we have made significant improvements to the conclusion in our revised manuscript.

In the revised paper, we not only highlight the applicational value and efficiency advantages of the DBO-Otsu algorithm in the field of image segmentation but also specifically mention the challenges and limitations encountered in practical applications. For example, we point out that directly applying the DBO-Otsu algorithm might not suffice for all scenarios when dealing with images of rubber tree diseases at different stages. Hence, we adopted a more nuanced approach, conducting morphological analyses post-initial segmentation and tailoring strategies to images at various disease stages.

Moreover, we mention in the conclusion that while the DBO-Otsu algorithm is able to perform tasks accurately in most instances, there is a slight decline in accuracy in images of intermediate disease stages. We emphasize the necessity of optimizing the algorithm in future research, including considering the integration of edge detection algorithms to eliminate irrelevant areas on the trunk, to enhance segmentation accuracy.

We believe that these additions and improvements provide a comprehensive presentation of the advantages of the DBO-Otsu algorithm, honestly reflect its limitations, and offer clear directions for future research.

• 2) Minor notes:

1. Comment: Please verify that acronym definitions are provided upon their initial use in the manuscript.

Response: We ensured that definitions for all acronyms are provided upon their initial appearance.

2. Comment: Please enhance the quality of figures. it would be beneficial to consider converting them into vector-based formats.

Response: We improved the quality of the figures and converted all the images in our paper into vector-based formats using Pace.

3. Comment: Please check some mistakes.

Response: We reviewed and corrected the errors in the manuscript.

We hope that these revisions meet the requirements of the reviewers and the editor, making our manuscript more suitable for publication. We look forward to your further guidance and feedback.

Sincerely,

Zhenjing Xie

Undergraduate Student

Hainan University

Haikou, 570228

China

20213005583@hainanu.edu.cn

Attachment

Submitted filename: Response to Reviewers.docx

Decision Letter 1

Khan Khan Bahadar Academic Editor

14 Dec 2023

PONE-D-23-30775R1Advancing Image Segmentation with DBO-Otsu: Addressing Rubber Tree Diseases through Enhanced Threshold TechniquesPLOS ONE

Dear Dr. Liu,

Thank you for submitting your manuscript to PLOS ONE. After careful consideration, we feel that it has merit but does not fully meet PLOS ONE's publication criteria as it currently stands. Therefore, we invite you to submit a revised version of the manuscript that addresses the points raised during the review process.

Please submit your revised manuscript by Jan 28 2024 11:59PM. If you will need more time than this to complete your revisions, please reply to this message or contact the journal office at plosone@plos.org. When you're ready to submit your revision, log on to https://www.editorialmanager.com/pone/ and select the 'Submissions Needing Revision' folder to locate your manuscript file.

Please include the following items when submitting your revised manuscript:

If you would like to make changes to your financial disclosure, please include your updated statement in your cover letter. Guidelines for resubmitting your figure files are available below the reviewer comments at the end of this letter.

If applicable, we recommend that you deposit your laboratory protocols in protocols.io to enhance the reproducibility of your results. Protocols.io assigns your protocol its own identifier (DOI) so that it can be cited independently in the future. For instructions see: https://journals.plos.org/plosone/s/submission-guidelines#loc-laboratory-protocols. Additionally, PLOS ONE offers an option for publishing peer-reviewed Lab Protocol articles, which describe protocols hosted on protocols.io. Read more information on sharing protocols at https://plos.org/protocols?utm_medium=editorial-email&utm_source=authorletters&utm_campaign=protocols.

We look forward to receiving your revised manuscript.

Kind regards,

Khan Bahadar Khan, Ph.D

Academic Editor

PLOS ONE

Journal Requirements:

Please review your reference list to ensure that it is complete and correct. If you have cited papers that have been retracted, please include the rationale for doing so in the manuscript text, or remove these references and replace them with relevant current references. Any changes to the reference list should be mentioned in the rebuttal letter that accompanies your revised manuscript. If you need to cite a retracted article, indicate the article's retracted status in the References list and also include a citation and full reference for the retraction notice.

[Note: HTML markup is below. Please do not edit.]

Reviewers' comments:

Reviewer's Responses to Questions

Comments to the Author

1. If the authors have adequately addressed your comments raised in a previous round of review and you feel that this manuscript is now acceptable for publication, you may indicate that here to bypass the "Comments to the Author" section, enter your conflict of interest statement in the "Confidential to Editor" section, and submit your "Accept" recommendation.

Reviewer #2: (No Response)

Reviewer #3: All comments have been addressed

***

2. Is the manuscript technically sound, and do the data support the conclusions?

The manuscript must describe a technically sound piece of scientific research with data that supports the conclusions. Experiments must have been conducted rigorously, with appropriate controls, replication, and sample sizes. The conclusions must be drawn appropriately based on the data presented.

Reviewer #2: (No Response)

Reviewer #3: Yes

***

3. Has the statistical analysis been performed appropriately and rigorously?

Reviewer #2: (No Response)

Reviewer #3: Yes

***

4. Have the authors made all data underlying the findings in their manuscript fully available?

The http://www.plosone.org/static/policies.action#sharing requires authors to make all data underlying the findings described in their manuscript fully available without restriction, with rare exception (please refer to the Data Availability Statement in the manuscript PDF file). The data should be provided as part of the manuscript or its supporting information, or deposited to a public repository. For example, in addition to summary statistics, the data points behind means, medians and variance measures should be available. If there are restrictions on publicly sharing data—e.g. participant privacy or use of data from a third party—those must be specified.

Reviewer #2: (No Response)

Reviewer #3: Yes

***

5. Is the manuscript presented in an intelligible fashion and written in standard English?

PLOS ONE does not copyedit accepted manuscripts, so the language in submitted articles must be clear, correct, and unambiguous. Any typographical or grammatical errors should be corrected at revision, so please note any specific errors here.

Reviewer #2: (No Response)

Reviewer #3: Yes

***

6. Review Comments to the Author

Please use the space provided to explain your answers to the questions above. You may also include additional comments for the author, including concerns about dual publication, research ethics, or publication ethics. (Please upload your review as an attachment if it exceeds 20,000 characters)

Reviewer #2: Please see the attached file.........................................................................

Reviewer #3: The introduction would benefit from a more detailed description of the Rubber Tree Tapping Panel Dryness (TPD) disease and its implications within the rubber industry, thereby providing essential context for the study. A succinct presentation of the research gap and how this work aims to bridge it is also recommended for clarifying the research's significance.

***

7. PLOS authors have the option to publish the peer review history of their article (https://journals.plos.org/plosone/s/editorial-and-peer-review-process#loc-peer-review-history). If published, this will include your full peer review and any attached files.

If you choose "no", your identity will remain anonymous but your review may still be made public.

Do you want your identity to be public for this peer review? For information about this choice, including consent withdrawal, please see our https://www.plos.org/privacy-policy.

Reviewer #2: No

Reviewer #3: No

***

[NOTE: If reviewer comments were submitted as an attachment file, they will be attached to this email and accessible via the submission site. Please log into your account, locate the manuscript record, and check for the action link "View Attachments". If this link does not appear, there are no attachment files.]

While revising your submission, please upload your figure files to the Preflight Analysis and Conversion Engine (PACE) digital diagnostic tool, https://pacev2.apexcovantage.com/. PACE helps ensure that figures meet PLOS requirements. To use PACE, you must first register as a user. Registration is free. Then, login and navigate to the UPLOAD tab, where you will find detailed instructions on how to use the tool. If you encounter any issues or have any questions when using PACE, please email PLOS at figures@plos.org. Please note that Supporting Information files do not need this step.

Attachment

Submitted filename: review-report-R1.pdf

Author response to Decision Letter 1

18 Dec 2023

Zhenjing Xie

Hainan University

Haikou, 570228

China

2023.11.28

Khan Bahadar Khan, Ph.D

Academic Editor

PLOS ONE

Dear Editor Khan Bahadar Khan,

Thank you and the reviewers for the review and recommendations on our manuscript (Title: "Advancing Image Segmentation with DBO-Otsu: Addressing Rubber Tree Diseases through Enhanced Threshold Techniques," Manuscript ID: PONE-D-23-30775). We have made detailed revisions based on the reviewer's feedback. Below is our response to the reviewer's comments:

Comment: The justification for selecting DBO over other MHs is still insufficient. It would be beneficial to emphasize the successful applications of DBO in the literature for addressing real-world problems. Additionally, please explicitly highlight the merits of DBO in comparison to other algorithms. Considering the problemdependent nature of metaheuristics and referring to the principles of the free lunch theorem would enhance the justification. Please include references.

Response: In response to your comment, Section 1.3 has been updated to include four recent publications featuring the application of the DBO optimization algorithm. These additions emphasize the suitability of the DBO algorithm for our research, particularly in terms of segmentation speed and accuracy, compared to other metaheuristic algorithms. Additionally, we have incorporated the 'No Free Lunch' theorem to acknowledge that no algorithm is universally perfect. This inclusion underscores the importance of selecting an algorithm that is most apt for the specific requirements of our study. Relevant references have been added to support these points.

• 2. Comment: It is difficult to discern the unique contributions of the proposed model. While the integration of a recent DBO with Otsu is good, it does not constitute a novel contribution. In this regard, I recommend highlighting the unique enhancement to the traditional Otsu (presented is section 3.1) explicitly in the list of contributions. Response: In light of your observation regarding the distinctiveness of our model's contributions, we have revised the final part of the Introduction to explicitly include the unique enhancements made to the traditional Otsu method in the list of contributions. Specifically, this update is reflected in the second point of the contributions section. This revision aims to clearly delineate our novel contribution, namely the integration of the recent DBO with Otsu, and its significance in advancing the methodology.
• 3. Comment: The authors state "Providing a theoretical analysis of the DBO algorithm", but the presented work is primarily experimental.

Response: In response to your comment on the theoretical analysis of the DBO algorithm, we have updated the contribution section by removing the reference to 'Providing a theoretical analysis of the DBO algorithm.' Acknowledging the primarily experimental nature of our work, we agree that mentioning a theoretical analysis could be misleading. This revision ensures that the contributions listed are in alignment with the experimental focus of our research.

4. Comment: Authors mentioned the use of an early convergence critera "During the iteration process, if the optimal threshold obtained is the same as the result of the first 5 iterations, the optimization process is considered to have converged and the iteration can be ended earlier, thus reducing the time consumption of the algorithm". It is unclear whether this approach is applied to other compared algorithms. Additionally, the rationale behind choosing 5 iterations as the threshold for convergence needs clarification.

Response: Regarding the early convergence criteria mentioned in our manuscript, we acknowledge that the theoretical basis of such criteria is detailed in the paper 'Effective Probabilistic Stopping Rules for Randomized Metaheuristics: GRASP Implementations.' However, it is important to note that this is not the focal point of our current study. The specific approach of terminating the iteration process after five identical results in our study was based on empirical experience and has minimal impact on the experimental outcomes. Consequently, we have decided to remove this content from the original manuscript. Furthermore, we clarify that this early termination method has not been applied to the other algorithms compared in our research, ensuring consistency in methodology across all tested approaches.

5. Comment: Results in Table 2 does not demonstrate the superiority of the proposed approach in most cases. Authors mentioned that "In evaluating average PSNR, FSIM, and SSIM scores, DBO-OTSU consistently ranked at or near the top..." However, it is not clear if a specific ranking method was employed to support this claim. If so, kindly present the rank for each method. Utilizing a widely used ranking method like Friedman rank would be beneficial.

Response: To address your concern regarding the demonstration of our approach's superiority in Table 2, we have conducted a comprehensive statistical analysis. We calculated the average rankings for each algorithm in terms of PSNR, FSIM, and SSIM scores. These rankings have been incorporated into Section 4.3 in a tabular format, accompanied by a brief explanation. This addition provides a clearer and more intuitive comparison among the algorithms, especially in scenarios where the differences in scores are not significantly pronounced. The use of average rankings offers a more structured way to assess the relative performance of each method, allowing for a fair and transparent comparison.

6. Comment: How do the authors justify the claim that, despite achieving better fitness scores, certain algorithms may not perform better in terms of other evaluation metrics like PSNR, FSIM, and SSIM?

Response: To justify our claim regarding the discrepancy between achieving better fitness scores and the performance in other evaluation metrics like PSNR, FSIM, and SSIM, we clarify that maximizing fitness, which is the inter-class variance in the mid-to-high threshold range, primarily focuses on enhancing the contrast between the foreground and background in grayscale. While this can improve segmentation effectiveness, it does not necessarily translate to higher overall image quality. PSNR, FSIM, and SSIM evaluate the segmented multi-threshold image against the original image. The application of multi-threshold Otsu's method might lead to loss of certain details in the image, particularly in high-contrast areas and regions with complex textures like tree barks, potentially lowering PSNR. FSIM considers image phase congruency and gradient magnitude to assess visual quality. However, the algorithm alters local features, especially at edges and textures, affecting the FSIM score. SSIM measures the structural, luminance, and contrast similarity between two images. The segmentation changes brought by the algorithm can alter local structures, thereby impacting the SSIM score.

However, this should not be construed as diminishing the relevance of PSNR, FSIM, and SSIM scores. These metrics provide a comprehensive assessment of the segmented multi-threshold images from a conventional perspective, complementing and corroborating our specialized evaluation approach.

7. Comment: Statistical analysis is important to judge the significance of the findings. In this comment I mean that Utilizing tests such as the Wilcoxon rank sum test would be beneficial to determine whether the differences presented in Table 2, such as those in PSNR (e.g., 8.762745 and 8.783505), are statistically significant or not."

Response: To address the need for statistical analysis to ascertain the significance of our findings, we have augmented Section 4.3 with a table presenting the results of the Wilcoxon signed-rank test. This test compares the DBO algorithm with other algorithms, using a predefined significance level of 0.1. The p-values obtained from this analysis have been duly recorded. The results demonstrate that the DBO algorithm shows statistically significant improvements in the PSNR and SSIM metrics. However, for the FSIM metric, the improvements were not found to be statistically significant. This additional analysis provides a robust statistical basis for validating the superiority of the DBO algorithm in specific aspects of our study.

8. Comment: Please verify that acronym definitions are provided upon their initial use in the manuscript and consistently use the acronym in subsequent parts.

Response: We have meticulously reviewed the entire manuscript in response to your comment regarding the use of acronyms. We have corrected the instances where acronyms were incorrectly used or not defined upon their initial use. This ensures that all acronyms are now properly introduced and consistently applied throughout the manuscript, enhancing its readability and clarity.

9. Comment: Please double check some mistakes

Response: In response to your comment regarding the presence of errors, we have thoroughly re-examined the manuscript and have identified and corrected the mistakes at the relevant locations. We appreciate your attention to detail and are committed to ensuring the accuracy and quality of our work.

10. Comment: Please enhance the quality of figures. (still need improvements, for instance the DBO flowchart .....)

Response: In response to your feedback on the quality of the figures, specifically mentioning the DBO flowchart, we have undertaken a comprehensive redesign of this figure to enhance its clarity and overall presentation. We hope the revised figure now meets your expectations in terms of visual quality and effectiveness in conveying the intended information.

11. Comment: Please review the order and placement of figures to ensure that captions are appropriately associated with the corresponding figures throughout the manuscript.

Response: In response to your comment on ensuring appropriate alignment of figure captions with their corresponding figures, we have conducted a thorough review of the entire manuscript. This review focused on verifying the order and placement of all figures to guarantee that each caption accurately reflects and is correctly associated with its respective figure. We have made necessary adjustments to ensure this consistency throughout the document.

We hope that these revisions meet the requirements of the reviewers and the editor, making our manuscript more suitable for publication. We look forward to your further guidance and feedback.

Sincerely,

Zhenjing Xie

Undergraduate Student

Hainan University

Haikou, 570228

China

20213005583@hainanu.edu.cn

Attachment

Submitted filename: Response to Reviewers.docx

Khan Khan Bahadar Academic Editor

Decision Letter 2

3 Jan 2024

Advancing Image Segmentation with DBO-Otsu: Addressing Rubber Tree Diseases through Enhanced Threshold Techniques

PONE-D-23-30775R2

Dear Dr. Liu,

We're pleased to inform you that your manuscript has been judged scientifically suitable for publication and will be formally accepted for publication once it meets all outstanding technical requirements.

Within one week, you'll receive an e-mail detailing the required amendments. When these have been addressed, you'll receive a formal acceptance letter and your manuscript will be scheduled for publication.

An invoice for payment will follow shortly after the formal acceptance. To ensure an efficient process, please log into Editorial Manager at http://www.editorialmanager.com/pone/, click the 'Update My Information' link at the top of the page, and double check that your user information is up-to-date. If you have any billing related questions, please contact our Author Billing department directly at authorbilling@plos.org.

If your institution or institutions have a press office, please notify them about your upcoming paper to help maximize its impact. If they'll be preparing press materials, please inform our press team as soon as possible -- no later than 48 hours after receiving the formal acceptance. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information, please contact onepress@plos.org.

Kind regards,

Khan Bahadar Khan, Ph.D

Academic Editor

PLOS ONE

Additional Editor Comments (optional):

Reviewers' comments:

Reviewer's Responses to Questions

Comments to the Author

1. If the authors have adequately addressed your comments raised in a previous round of review and you feel that this manuscript is now acceptable for publication, you may indicate that here to bypass the "Comments to the Author" section, enter your conflict of interest statement in the "Confidential to Editor" section, and submit your "Accept" recommendation.

Reviewer #2: All comments have been addressed

Reviewer #4: All comments have been addressed

***

2. Is the manuscript technically sound, and do the data support the conclusions?

The manuscript must describe a technically sound piece of scientific research with data that supports the conclusions. Experiments must have been conducted rigorously, with appropriate controls, replication, and sample sizes. The conclusions must be drawn appropriately based on the data presented.

Reviewer #2: Yes

Reviewer #4: Yes

***

3. Has the statistical analysis been performed appropriately and rigorously?

Reviewer #2: Yes

Reviewer #4: Yes

***

4. Have the authors made all data underlying the findings in their manuscript fully available?

The http://www.plosone.org/static/policies.action#sharing requires authors to make all data underlying the findings described in their manuscript fully available without restriction, with rare exception (please refer to the Data Availability Statement in the manuscript PDF file). The data should be provided as part of the manuscript or its supporting information, or deposited to a public repository. For example, in addition to summary statistics, the data points behind means, medians and variance measures should be available. If there are restrictions on publicly sharing data—e.g. participant privacy or use of data from a third party—those must be specified.

Reviewer #2: Yes

Reviewer #4: Yes

***

5. Is the manuscript presented in an intelligible fashion and written in standard English?

PLOS ONE does not copyedit accepted manuscripts, so the language in submitted articles must be clear, correct, and unambiguous. Any typographical or grammatical errors should be corrected at revision, so please note any specific errors here.

Reviewer #2: Yes

Reviewer #4: Yes

***

6. Review Comments to the Author

Please use the space provided to explain your answers to the questions above. You may also include additional comments for the author, including concerns about dual publication, research ethics, or publication ethics. (Please upload your review as an attachment if it exceeds 20,000 characters)

Reviewer #2: All comments have been adequately addressed

.................................................................................

Reviewer #4: The paper in its present second submission looks like fully ready to be considered for publication. All past comments have been fully addressed and more explanations have been added to the discussion. The results section has been appropriately widen as to include more information about the algorithm performance itself and its comparison to other algorithms is now clearer and more feasible.

***

7. PLOS authors have the option to publish the peer review history of their article (https://journals.plos.org/plosone/s/editorial-and-peer-review-process#loc-peer-review-history). If published, this will include your full peer review and any attached files.

If you choose "no", your identity will remain anonymous but your review may still be made public.

Do you want your identity to be public for this peer review? For information about this choice, including consent withdrawal, please see our https://www.plos.org/privacy-policy.

Reviewer #2: No

Reviewer #4:  Yes:  Marco Perez-Cisneros

***

Khan Khan Bahadar Academic Editor

Acceptance letter

25 Jan 2024

PONE-D-23-30775R2

PLOS ONE

Dear Dr. Liu,

I'm pleased to inform you that your manuscript has been deemed suitable for publication in PLOS ONE. Congratulations! Your manuscript is now being handed over to our production team.

At this stage, our production department will prepare your paper for publication. This includes ensuring the following:

* All references, tables, and figures are properly cited

* All relevant supporting information is included in the manuscript submission,

* There are no issues that prevent the paper from being properly typeset

If revisions are needed, the production department will contact you directly to resolve them. If no revisions are needed, you will receive an email when the publication date has been set. At this time, we do not offer pre-publication proofs to authors during production of the accepted work. Please keep in mind that we are working through a large volume of accepted articles, so please give us a few weeks to review your paper and let you know the next and final steps.

Lastly, if your institution or institutions have a press office, please let them know about your upcoming paper now to help maximize its impact. If they'll be preparing press materials, please inform our press team within the next 48 hours. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information, please contact onepress@plos.org.

If we can help with anything else, please email us at customercare@plos.org.

Thank you for submitting your work to PLOS ONE and supporting open access.

Kind regards,

PLOS ONE Editorial Office Staff

on behalf of

Dr. Khan Bahadar Khan

Academic Editor

PLOS ONE

1 The authors have declared that no competing interests exist. Habib M. A. H., Ismail M. N.Hevea brasiliensis latex proteomics: a review of analytical methods and the way forward. Journal of Plant Research 134, 43–53 ((2021)). doi: 10.1007/s10265-020-01231-x, 33108557 2 Herlinawati E.et al. Dynamic analysis of Tapping Panel Dryness in Hevea brasiliensis reveals new insights on this physiological syndrome affecting latex production. Heliyon 8, ((2022)). 3 Y. Nakanoet al. Transcriptome analysis of Para rubber tree (H. brasiliensis) seedlings under ethylene stimulation. Bmc Plant Biology 21, ((2021)). 4 Ding Z. H., Fu L. L., Tan D. G., Sun X. P., Zhang J. M.An integrative transcriptomic and genomic analysis reveals novel insights into the hub genes and regulatory networks associated with rubber synthesis in H. brasiliensis. Industrial Crops and Products 153, ((2020)). doi: 10.1016/j.indcrop.2020.112562 5 Li Zhaotong. Screening and Identification of Virus Related to Tapping Panel Dryness (TDP) in Hevea Brasiliensis. [Master's thesis, Hainan University], (2020). 6 YANG Hong, WANG Lifeng, DAI Longjun, GUO Bingbing. Effects of Tapping Panel Dryness on Mitochondrial Ultrastructure and ROS Metabolism in Barks of Rubber Tree. Bulletin of Botanical Research 43, 69–75 ((2023)). 7 Liu H.et al. Genome-wide identification of lncRNAs, miRNAs, mRNAs and their regulatory networks involved in tapping panel dryness in rubber tree (Hevea brasiliensis). Tree Physiology 42, 629–645 ((2022)). doi: 10.1093/treephys/tpab120 8 Ashraf M. A., Tariq H. K., Hu X. W., Khan J., Zou Z.Computational Biology and Machine Learning Approaches Identify Rubber Tree (Hevea brasiliensis Muell. Arg.) Genome Encoded MicroRNAs Targeting Rubber Tree Virus 1. Applied Sciences-Basel 12, ((2022)). 9 Nie Z. Y.et al. Downregulation of HbFPS1 affects rubber biosynthesis of Hevea brasiliensis suffering from tapping panel dryness. Plant Journal 113, 504–520 ((2023)). doi: 10.1111/tpj.16063, 36524729 Leclercq J.et al. Post-transcriptional regulation of several biological processes involved in latex production in Hevea brasiliensis. Peerj 8, ((2020)). Yue Y.et al. Bark transcriptome analyses reveals molecular mechanisms involved in tapping panel dryness occurrence and development in rubber tree (Hevea brasiliensis). Gene 892, 147894 ((2024)). doi: 10.1016/j.gene.2023.147894 Yuan K., He J., Hu Y. Y., Feng C. T., Wang Z. H.The variation of reactive oxygen species scavenging enzymes and related gene expressions during occurrence and recovery of rubber tree tapping panel dryness. Journal of Rubber Research 24, 391–402 ((2021)). doi: 10.1007/s42464-021-00106-7 Yu Y.et al. Techniques and Challenges of Image Segmentation: A Review. Electronics 12, ((2023)). Singh N., Bhandari A. K., Kumar I. V.Fusion-based contextually selected 3D Otsu thresholding for image segmentation. Multimedia Tools and Applications 80, 19399–19420 ((2021)). doi: 10.1007/s11042-021-10706-5 Wu J., Wang X., Wei T., Fang C.Full-parameter adaptive fuzzy clustering for noise image segmentation based on non-local and local spatial information. Computer Vision and Image Understanding, 103765 ((2023)). doi: 10.1016/j.cviu.2023.103765 Xu R.Fuzzy C-means Clustering Image Segmentation Algorithm Based on Hidden Markov Model. Mobile Networks & Applications 27, 946–954 ((2022)). doi: 10.1007/s11036-022-01917-7 Faragallah O. S., El-Hoseny H. M., El-sayed H. S.Efficient brain tumor segmentation using Otsu and K-means clustering in homomorphic transform. Biomedical Signal Processing and Control 84, ((2023)). doi: 10.1016/j.bspc.2023.104712 Guo L. G., Wu S. T.FPGA Implementation of a Real-Time Edge Detection System Based on an Improved Canny Algorithm. Applied Sciences-Basel 13, ((2023)). Chen C., Kong H., Wu B.Edge detection of remote sensing image based on Grünwald-Letnikov fractional difference and Otsu threshold. Electronic Research Archive 31, 1287–1302 ((2023)). doi: 10.3934/era.2023066 Chen F., Zhang A. H., Balzter H., Ren P., Zhou H. Y.Oil Spill SAR Image Segmentation via Probability Distribution Modeling. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing 15, 533–554 ((2022)). doi: 10.1109/JSTARS.2021.3136089 Chen C.et al. Deep learning for cardiac image segmentation: a review. Frontiers in Cardiovascular Medicine7, 25 ((2020)). Minaee S.et al. Image Segmentation Using Deep Learning: A Survey. IEEE Transactions on Pattern Analysis and Machine Intelligence 44, 3523–3542 ((2022)). Xu R. L., Zhao S. Q., Ke Y. H.A Simple Phenology-Based Vegetation Index for Mapping Invasive Spartina Alterniflora Using Google Earth Engine. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing 14, 190–201 ((2021)). doi: 10.1109/JSTARS.2020.3038648 Prabaharan L., Raghunathan A.An improved convolutional neural network for abnormality detection and segmentation from human sperm images. Journal of Ambient Intelligence and Humanized Computing 12, 3341–3352 ((2021)). doi: 10.1007/s12652-020-02773-7 Yan Q.et al. 3D Medical image segmentation using parallel transformers. Pattern Recognition 138, 109432 ((2023)). doi: 10.1016/j.patcog.2023.109432 Qian L., Huang H., Xia X., Li Y., Zhou X.Automatic segmentation method using FCN with multi-scale dilated convolution for medical ultrasound image. The Visual Computer, ((2022)). Pham M. V., Ha Y. S., Kim Y. T.Automatic detection and measurement of ground crack propagation using deep learning networks and an image processing technique. Measurement 215, ((2023)). doi: 10.1016/j.measurement.2023.112832 Fu Y. C.et al. Image Segmentation of Cabin Assembly Scene Based on Improved RGB-D Mask R-CNN. IEEE Transactions on Instrumentation and Measurement 71, ((2022)). doi: 10.1109/TIM.2022.3145388 Geng Q., Yan H. F.Image Segmentation under the Optimization Algorithm of Krill Swarm and Machine Learning. Computational Intelligence and Neuroscience 2022, ((2022)). doi: 10.1155/2022/8771650, 35371201 Pare S., Kumar A., Singh G. K., Bajaj V.Image Segmentation Using Multilevel Thresholding: A Research Review. Iranian Journal of Science and Technology-Transactions of Electrical Engineering 44, 1–29 ((2020)). doi: 10.1007/s40998-019-00251-1 Wang S. X., Fan J. L.Simplified expression and recursive algorithm of multi-threshold Tsallis entropy. Expert Systems with Applications 237, ((2024)). doi: 10.1016/j.eswa.2023.121690 Sharma S. R.et al. Hybrid Multilevel Thresholding Image Segmentation Approach for Brain MRI. Diagnostics 13, ((2023)). doi: 10.3390/diagnostics13050925, 36900074 Lei B., Fan J. L.Adaptive granulation Renyi rough entropy image thresholding method with nested optimization. Expert Systems with Applications 203, ((2022)). doi: 10.1016/j.eswa.2022.117378 A. Rahim. in Laser Science. Optica Publishing Group, 2022, pp. JW4B. 72. Xing J. W., Yang P., Qingge L. G.Robust 2D Otsu's Algorithm for Uneven Illumination Image Segmentation. Computational Intelligence and Neuroscience 2020, ((2020)). doi: 10.1155/2020/5047976, 32849864 Bhandari A. K., Ghosh A., Kumar I. V.A local contrast fusion based 3D Otsu algorithm for multilevel image segmentation. IEEE-Caa Journal of Automatica Sinica 7, 200–213 ((2020)). doi: 10.1109/JAS.2019.1911843 Bhandari A. K., Singh A., Kumar I. V.Spatial Context Energy Curve-Based Multilevel 3-D Otsu Algorithm for Image Segmentation. IEEE Transactions on Systems Man Cybernetics-Systems 51, 2760–2773 ((2021)). doi: 10.1109/TSMC.2019.2916876 Chen M. Y., Zhang Z. Y., Wu H., Xie S. L., Wang H.Otsu-Kmeans gravity-based multi-spots center extraction method for microlens array imaging system. Optics and Lasers in Engineering 152, ((2022)). doi: 10.1016/j.optlaseng.2022.106968 Ma J. C., Cheng X. D.Fast segmentation algorithm of PCB image using 2D Otsu improved by adaptive genetic algorithm and integral image. Journal of Real-Time Image Processing 20, ((2023)). doi: 10.1007/s11554-023-01272-0 Xiao L. Y., Fan C. D., Ouyang H. L., Abate A. F., Wan S. H.Adaptive trapezoid region intercept histogram based Otsu method for brain MR image segmentation. Journal of Ambient Intelligence and Humanized Computing 13, 2161–2176 ((2022)). doi: 10.1007/s12652-021-02976-6 Dhal K. G., Das A., Ray S., Galvez J., Das S.Nature-Inspired Optimization Algorithms and Their Application in Multi-Thresholding Image Segmentation. Archives of Computational Methods in Engineering 27, 855–888 ((2020)). doi: 10.1007/s11831-019-09334-y Saeidifar M., Yazdi M., Zolghadrasli A.Performance Improvement in Brain Tumor Detection in MRI Images Using a Combination of Evolutionary Algorithms and Active Contour Method. Journal of Digital Imaging 34, 1209–1224 ((2021)). doi: 10.1007/s10278-021-00514-6 Chai R. S.Otsu's Image Segmentation Algorithm with Memory-Based Fruit Fly Optimization Algorithm. Complexity 2021, ((2021)). doi: 10.1155/2021/5564690 Ning G. Y.Two-dimensional Otsu multi-threshold image segmentation based on hybrid whale optimization algorithm. Multimedia Tools and Applications 82, 15007–15026 ((2023)). doi: 10.1007/s11042-022-14041-1 Abdel-Basset M., Mohamed R., AbdelAziz N. M., Abouhawwash M.HWOA: A hybrid whale optimization algorithm with a novel local minima avoidance method for multi-level thresholding color image segmentation. Expert Systems with Applications 190, 116145 ((2022)). doi: 10.1016/j.eswa.2021.116145 Ma G. Y., Yue X. F.An improved whale optimization algorithm based on multilevel threshold image segmentation using the Otsu method. Engineering Applications of Artificial Intelligence 113, ((2022)). doi: 10.1016/j.engappai.2022.104960 Nadimi-Shahraki M. H., Zamani H., Varzaneh Z., Mirjalili S.A Systematic Review of the Whale Optimization Algorithm: Theoretical Foundation, Improvements, and Hybridizations. Archives of Computational Methods in Engineering, ((2023)). doi: 10.1007/s11831-023-09928-7, 37359740 Chen C. C., Wang X. C., Heidari A. A., Yu H. L., Chen H. L.Multi-Threshold Image Segmentation of Maize Diseases Based on Elite Comprehensive Particle Swarm Optimization and Otsu. Frontiers in Plant Science 12, ((2021)). doi: 10.3389/fpls.2021.789911, 34966405 Wunnava A., Naik M. K., Panda R., Jena B., Abraham A.An adaptive Harris hawks optimization technique for two dimensional grey gradient based multilevel image thresholding. Applied Soft Computing 95, ((2020)). doi: 10.1016/j.asoc.2020.106526 Qu X. F.et al. Fast detection of dam zone boundary based on Otsu thresholding optimized by enhanced harris hawks optimization. Plos One 18, ((2023)). doi: 10.1371/journal.pone.0271692, 36745651 Wunnava A., Naik M. K., Panda R., Jena B., Abraham A.A differential evolutionary adaptive Harris hawks optimization for two dimensional practical Masi entropy-based multilevel image thresholding. Journal of King Saud University-Computer and Information Sciences 34, 3011–3024 ((2022)). doi: 10.1016/j.jksuci.2020.05.001 Naik M. K., Panda R., Wunnava A., Jena B., Abraham A.A leader Harris hawks optimization for 2-D Masi entropy-based multilevel image thresholding. Multimedia Tools and Applications 80, 35543–35583 ((2021)). doi: 10.1007/s11042-020-10467-7 Rodríguez-Esparza E.et al. An efficient Harris hawks-inspired image segmentation method. Expert Systems with Applications 155, ((2020)). Mousavirad S. J., Schaefer G., Zhou H. Y., Moghadam M. H.How effective are current population-based metaheuristic algorithms for variance-based multi-level image thresholding? Knowledge-Based Systems 272, ((2023)). doi: 10.1016/j.knosys.2023.110587 Liu L.et al. Ant colony optimization with Cauchy and greedy Levy mutations for multilevel COVID 19 X-ray image segmentation. Computers in Biology and Medicine 136, ((2021)). Abualigah L., Diabat A., Sumari P., Gandomi A. H.A Novel Evolutionary Arithmetic Optimization Algorithm for Multilevel Thresholding Segmentation of COVID-19 CT Images. Processes 9, ((2021)). Sabha M., Thaher T., Emam M. M.Cooperative Swarm Intelligence Algorithms for Adaptive Multilevel Thresholding Segmentation of COVID-19 CT-Scan Images. Journal of Universal Computer Science 29, 759–804 ((2023)). doi: 10.3897/jucs.93498 Su H.et al. Multilevel threshold image segmentation for COVID-19 chest radiography: A framework using horizontal and vertical multiverse optimization. Computers in Biology and Medicine 146, ((2022)). Chen J. C.et al. Dynamic mechanism-assisted artificial bee colony optimization for image segmentation of COVID-19 chest X-ray. Displays 79, ((2023)). doi: 10.1016/j.displa.2023.102485 Emam M. M., Houssein E. H., Ghoniem R. M.A modified reptile search algorithm for global optimization and image segmentation: Case study brain MRI images. Computers in Biology and Medicine 152, ((2023)). doi: 10.1016/j.compbiomed.2022.106404, 36521356 Chen Y.et al. Multi-threshold image segmentation using a multi-strategy shuffled frog leaping algorithm. Expert Systems with Applications 194, ((2022)). Abdel-Basset M., Chang V., Mohamed R.A novel equilibrium optimization algorithm for multi-thresholding image segmentation problems. Neural Computing & Applications 33, 10685–10718 ((2021)). doi: 10.1007/s00521-020-04820-y Liu Y. Y., Sun J. H., Yu H. Y., Wang Y. Y., Zhou X. K.An Improved Grey Wolf Optimizer Based on Differential Evolution and Otsu Algorithm. Applied Sciences-Basel 10, ((2020)). Huang C. Y., Li X. R., Wen Y. L.AN Otsu image segmentation based on fruitfly optimization algorithm. Alexandria Engineering Journal 60, 183–188 ((2021)). doi: 10.1016/j.aej.2020.06.054 Houssein E. H., Abdelkareem D. A., Emam M. M., Hameed M. A., Younan M.An efficient image segmentation method for skin cancer imaging using improved golden jackal optimization algorithm. Computers in Biology and Medicine 149, ((2022)). doi: 10.1016/j.compbiomed.2022.106075, 36115303 Otair M.et al. Adapted arithmetic optimization algorithm for multi-level thresholding image segmentation: a case study of chest x-ray images. Multimedia Tools and Applications, ((2023)). Rather S. A., Das S.Levy Flight and Chaos Theory-Based Gravitational Search Algorithm for Image Segmentation. Mathematics 11, ((2023)). doi: 10.3390/math11183913 Liu Q. X., Li N., Jia H. M., Qi Q., Abualigah L.A chimp-inspired remora optimization algorithm for multilevel thresholding image segmentation using cross entropy. Artificial Intelligence Review 56, 159–216 ((2023)). doi: 10.1007/s10462-023-10498-0 Alhassan A. M., I. Alzheimer's Dis Neuroimaging, L. Australian Imaging BiomarkersEnhanced Fuzzy Elephant Herding Optimization-Based Otsu Segmentation and Deep Learning for Alzheimer's Disease Diagnosis. Mathematics 10, ((2022)). doi: 10.3390/math10081259 Zhu S. Y., Tian Y. J.Shape robustness in style enhanced cross domain semantic segmentation. Pattern Recognition 135, ((2023)). doi: 10.1016/j.patcog.2022.109143 J. Xue, B. Shen. in The Journal of Supercomputing. 2022, vol. 79, chap. 7305, pp. 7305–7336. Zhang R. Z., Zhu Y. J.Predicting the Mechanical Properties of Heat-Treated Woods Using Optimization-Algorithm-Based BPNN. Forests 14, ((2023)). Yuan Q. M., Wu L., Huang Y. B., Guo Z. W., Li N.Water-Body Detection From Spaceborne SAR Images With DBO-CNN. Ieee Geoscience and Remote Sensing Letters 20, ((2023)). doi: 10.1109/LGRS.2023.3325939 Xie Y. P., Zeng H. B., Yang K. J., Yuan Q. M., Yang C.Water-Body Detection in Sentinel-1 SAR Images with DK-CO Network. Electronics 12, ((2023)). doi: 10.3390/electronics12143163 Alamgeer M., Alruwais N., Alshahrani H. M., Mohamed A., Assiri M.Dung Beetle Optimization with Deep Feature Fusion Model for Lung Cancer Detection and Classification. Cancers 15, ((2023)). Zhao M. X., Jiang H., Chen Q. S.Identification of procymidone in rapeseed oils based on olfactory visualization technology. Microchemical Journal 193, ((2023)). doi: 10.1016/j.microc.2023.109055 Li C. H., Lee C.Minimum cross entropy thresholding. Pattern recognition 26, 617–625 ((1993)). doi: 10.1016/0031-3203(93)90115-D Huang D. Y., Wang C. H.Optimal multi-level thresholding using a two-stage Otsu optimization approach. Pattern Recognition Letters 30, 275–284 ((2009)). doi: 10.1016/j.patrec.2008.10.003 Ostu N.A threshold selection method from gray-level histograms. IEEE Trans SMC 9, 62 ((1979)). Xue J., Shen B.A novel swarm intelligence optimization approach: sparrow search algorithm. Systems science & control engineering 8, 22–34 ((2020)). doi: 10.1080/21642583.2019.1708830 Mirjalili S., Lewis A.The Whale Optimization Algorithm. Advances in Engineering Software 95, 51–67 ((2016)). doi: 10.1016/j.advengsoft.2016.01.008 Braik M., Hammouri A., Atwan J., Al-Betar M. A., Awadallah M. A.White Shark Optimizer: A novel bio-inspired meta-heuristic algorithm for global optimization problems. Knowledge-Based Systems 243, 108457 ((2022)). doi: 10.1016/j.knosys.2022.108457 Mirjalili S., Mirjalili S. M., Lewis A.Grey Wolf Optimizer. Advances in Engineering Software 69, 46–61 ((2014)). doi: 10.1016/j.advengsoft.2013.12.007 Zhao W., Wang L., Mirjalili S.Artificial hummingbird algorithm: A new bio-inspired optimizer with its engineering applications. Computer Methods in Applied Mechanics and Engineering 388, 114194 ((2022)). doi: 10.1016/j.cma.2021.114194 Braik M. S.Chameleon Swarm Algorithm: A bio-inspired optimizer for solving engineering design problems. Expert Systems with Applications 174, 114685 ((2021)). doi: 10.1016/j.eswa.2021.114685 Jie Zhang, Mohan Zhang, Shute Li; Haihua Xing, Yongna Liu. Extraction of Cut Marks and Identification of TPD of Rubber Tree Based on Image Processing. Computer Software and Application of Computer 36, 259 ((2019)).

By Zhenjing Xie; Jinran Wu; Weirui Tang and Yongna Liu

Reported by Author; Author; Author; Author