蔬菜Issue(8)：37-57,21.DOI:10.26924/j.cnki.sczz.2026.8004
基于深度学习的作物叶部病害识别技术研究进展
Research Progress of Crop Leaf Disease Recognition Technology Based on Deep Learning
摘要
Abstract
Accurate identification of crop foliar diseases is a key technical link in smart agriculture for disease prevention and control.The core challenges lie in visual complexities such as small lesion scales,diverse morphological patterns,complex backgrounds,and significant phenotypic variations.Deep learning,with its data-driven feature learning capabilities,has become the mainstream approach for intelligent crop foliar disease identification.This paper systematically reviews the research progress of three major deep learning paradigms including Convolutional Neural Network(CNN),Transformer model,and CNN-Transformer hybrid model in the field of crop foliar disease recognition.From the perspectives of three core visual tasks such as disease classification,lesion target detection,and lesion semantic segmentation,this paper provides an in-depth analysis of the modeling principles,technical advantages,applicable scenarios,and performance boundaries of each model type.Research indicates that CNN possesses structural advantages for local texture modeling,while Transformers compensate for CNN's deficiencies in global context modeling through self-attention mechanisms.CNN-Transformer hybrid model,through local-global collaborative mechanisms,demonstrates stronger robustness and recognition accuracy in complex field scenarios.On this basis,this paper analyzes the core bottlenecks of current disease recognition technologies,such as scarce labeled data,insufficient model generalization and limitations in practical deployment.Meanwhile,five major research directions for future development are proposed from the perspectives of adaptive multi-scale modeling,in-depth architecture fusion,few-shot learning,lightweight deployment and interpretable agricultural AI construction,so as to provide a systematic reference for algorithm optimization,scene adaptation and practical application of intelligent recognition of crop leaf diseases.关键词
作物病害识别/深度学习/卷积神经网络/Transformer模型/混合模型/智慧农业Key words
crop disease recognition/deep learning/Convolutional Neural Networks(CNN)/Transformer model/hybrid model/smart agriculture分类
农业科技引用本文复制引用
李立博,王晴雪,王紫凡,王建春,王春山..基于深度学习的作物叶部病害识别技术研究进展[J].蔬菜,2026,(8):37-57,21.基金项目
国家自然科学基金项目(62472012) （62472012）
石家庄市骨干人才(科技平台)个人项目(2687900206A). （科技平台）
