基于改进YOLOv7的肥城桃病虫害识别方法  

Identification Method of Feicheng Peach Diseases and Pests Based on Improved YOLOv7

作　　者：刘鹏 周鑫[2] 孙博 陈维康 王志军[1] Liu Peng;Zhou Xin;Sun Bo;Chen Weikang;Wang Zhijun(College of Information Science and Engineering,Shandong Agricultural University,Taian 271018,China;Shandong Taian No.1 Middle School,Taian 271000,China)

机构地区：[1]山东农业大学信息科学与工程学院,山东泰安271018 [2]山东省泰安第一中学,山东泰安271000

出　　处：《山东农业科学》2024年第8期150-157,共8页Shandong Agricultural Sciences

基　　金：山东省重大科技创新工程项目“现代果园智慧种植装备与大数据平台研发及示范应用”(2019JZZY010706)。

摘　　要：为解决肥城桃病虫害特征小以及不同病斑表征相似导致的难以精准识别的问题,以山东省肥城市肥城桃种植基地为样本采集点,构建包含细菌穿孔病、褐斑穿孔病、潜隐黄化病、桃小食心虫、红颈天牛、流胶病6种桃病虫害的数据集;针对样本分布特点,引入Mixup、Cutout、高斯模糊等多种方法进行数据增强,以提升模型对小病斑特征的检测;以YOLOv7模型作为骨干网络,加入Ghost模块进行瘦身以降低模型冗余特征,构建基于CBAM注意力机制和加权双向特征金字塔网络(BiFPN)的多尺度神经网络模型,增强小病斑的多尺度融合,从而提高模型的泛化能力。经实验验证,改进后的模型对上述6种病虫害的识别精度均值(mAP)达到93.2%。表明改进后的模型能够实现对病虫害的有效识别,可为肥城桃病虫害的早期预警和防治提供一定的技术支撑。In order to solve the difficulties in accurate identification of diseases and pests in Feicheng peach due to small size of features and morphological similarity of disease spots,the Feicheng peach planting base in Shandong Province was used as sample collection site,and a dataset was constructed containing six peach diseases and pests including bacterial perforation disease,brown spot perforation disease,hidden yellowing disease,peach small heart borer,red necked beetle and gummosis.Based on distribution characteristics of samples,various methods such as Mixup,Cutout and Gaussian Blur were introduced for data augmentation to improve the model’s detection accuracy of small disease spot features.Using YOLOv7 model as backbone network,Ghost module was added for slimming to reduce redundant features of the model,and multi-scale neural network model was constructed based on CBAM attention mechanism and weighted bidirectional feature pyramid network(BiFPN)to enhance multi-scale fusion of small disease spots and improve the generalization ability of the model.After experimental verification,the improved model achieved the mean average precision(mAP)of 93.2% for the six diseases and insects mentioned above,which indicated that it could identify pests and diseases effectively,and provide technical supports for early warning and prevention of diseases and pests in Feicheng peach.

关 键 词：肥城桃 病虫害识别 YOLOv7模型 深度学习 卷积神经网络 

分 类 号：S126[农业科学—农业基础科学] S436.621