基于深度学习的桑叶病害识别方法研究  

Research on mulberry leaf disease recognition method based on deep learning

作　　者：叶晖 项东晖 曾松伟[3] Ye Hui;Xiang Donghui;Zeng Songwei(School of Mathematics and Computer Science,Zhejiang A&F University,Hangzhou 311300,China;Zhejiang Yigangtong E-commerce Co.,Ltd.,Ningbo 315200,China;School of Optoelectromechanical Engineering,Zhejiang A&F University,Hangzhou 311300,China)

机构地区：[1]浙江农林大学数学与计算机科学学院,浙江杭州311300 [2]浙江易港通电子商务有限公司,浙江宁波315200 [3]浙江农林大学光机电工程学院,浙江杭州311300

出　　处：《电子技术应用》2025年第3期70-76,共7页Application of Electronic Technique

摘　　要：为提高桑叶病害检测精度,实现将模型方便快速部署到移动端,针对自然环境下桑叶病害病斑小、背景复杂等问题,以YOLOv8为基线模型进行改进,提出了一种YOLOv8-Evo的桑叶病害识别算法。首先在Backbone模块中加入了可变形卷积模块从而更灵活地捕捉病害的细节和形状,其次在Neck模块中增加了CBAM(Convolutional Block Attention Module)注意力机制,发掘图像中的关键特征和区域,最后在18849张桑叶病害数据集上进行验证,相较YOLOv8s模型,YOLOv8-Evo的识别精度提高2.4%,召回率提高1.5%,mAP50提高1%,mAP50-95提高0.7%,实验证明改进的YOLOv8-Evo模型为桑叶病害识别的自动化提供了理论依据与技术支持。To improve the accuracy of mulberry leaf disease detection and enable convenient and rapid deployment of models on mobile devices,an improved version of the YOLOv8 model,named YOLOv8-Evo,is proposed to address issues such as small lesion spots and complex backgrounds in natural environments.The algorithm introduces a deformable convolution module within the Backbone to capture disease details and shapes more flexibly.Additionally,a Convolutional Block Attention Module(CBAM)is incorporated into the Neck to highlight key features and regions in the image.After validation on a dataset of 18849 mulberry leaf disease images,the YOLOv8-Evo model demonstrates a 2.4%increase in precision,a 1.5% increase in recall rate,a 1% improvement in mAP50,and a 0.7% improvement in mAP50-95 compared to the YOLOv8s model.These results provide both theoretical support and technical backing for the automation of mulberry leaf disease identification.

关 键 词：桑叶病害 YOLOv8 目标检测 模型改进 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]