Research on Pixel-level Segmentation Method for Leaf Diseases of Multiple Crops
WANG Congsheng
LI Qingyang
QIU Xiurong
Abstract：Precise pixel-level segmentation is a key technology for achieving automated diagnosis and quantitative assessment of crop leaf diseases.However,the generalization performance of existing models across crops and diseases has not been fully verified.In this study,the public PlantDoc dataset was used as the experimental material to systematically evaluate the generalization performance of the mainstream deep learning models U-Net and DeepLabV3+in the pixel-level segmentation task of multiple crop leaf diseases.The results showed that the U-Net model showed excellent segmentation performance on the PlantDoc dataset,with the average intersection over union,average pixel accuracy and accuracy were 77.20%,84.50%and 93.64%respectively,which were better than the corresponding indicators of the DeepLabV3+model,which were 55.63%,62.77%and 86.41%respectively.The research results provide a model selection basis for agricultural intelligent diagnosis system.
Keywords：CropU-Net modelDeepLabV3+modelLeaf diseasesImage segmentation
Publication Date：2026-02-10
Online Publishing Date：2026-09-12(First online date of this platform, not the publication date of the document)
Pages：4( 70-73 )