Real-time crop recognition in transplanted fields with prominent weed growth: a visual-attention-based approach
In: oai:zenodo.org:7954695; (2020)
Online
Elektronische Ressource
Zugriff:
Real-time crop recognition in transplanted fields with prominent weed growth: a visual-attention-based approach In this work, a novel strategy for crop recognition in fields with high weed infestation levels is devised based on visual attention mechanism. It detects crop plants based on their saliency in the images, which is invariant to species of crops and weeds, and thus more generalized than features like color, shape and texture. The saliency of crop plants over surrounding weeds has been demonstrated in our experiment that the CNN model trained with a general saliency detection dataset (MSRA-B, containing no field images) can detect crop plants in general. A deep CNN framework is proposed to detect salient regions (crop plants) in field images based on the DSS architecture proposed by Hou et al (see reference [28] in our paper). The network possesses side-output structures with short connections for extracting multi-scale features, while taking advantage of Adaptive Affinity Fields to improve the segmentation at boundaries and fine structures. It is lightweight, efficient, and able to accurately segment crop plants from weedy backgrounds. A field image dataset, CWF-788 (CWF is short for crop in weedy field), is created to train and evaluate the proposed method. It contains 788 images captured from cauliflower fields with high weed pressure. High-quality pixel-wise annotated label images are provided with the dataset. This dataset can be used for future researches and fair comparisons. Datasets We provide our CWF-788 dataset with two different resolutions,IMAGE400x300 with the resolution of 400x300 and IMAGE512x384 with the resolution of 512x384. Usage Please install Tensorflow and required packages first. Download our code. git clone https://github.com/ZhangXG001/Real-Time-Crop-Recognition.git Create .csv files of dataset. Replace the line 13 path = './dataset/'+ dirname +'/' of csv_generator.py with path = './IMAGE400x300/'+ dirname +'/',if you use the dataset IMAGE400x300. Replace the line 13 path = ...
Titel: |
Real-time crop recognition in transplanted fields with prominent weed growth: a visual-attention-based approach
|
---|---|
Autor/in / Beteiligte Person: | Nan,Li ; Xiaoguang, Zhang ; Chunlong, Zhang ; Huiwen, Zuo ; Zhe, Sun ; Xinyu, Wu |
Link: | |
Quelle: | oai:zenodo.org:7954695; (2020) |
Veröffentlichung: | 2020 |
Medientyp: | Elektronische Ressource |
DOI: | 10.5281/zenodo.7954695 |
Schlagwort: |
|
Sonstiges: |
|