基于改进YOLOv4的葡萄叶片病害检测  

Detection of grape leaf diseases based on improved YOLOv4

作　　者：雷建云[1] 叶莎 夏梦[1] 郑禄[1] 邹金林 LEI Jianyun;YE Sha;XIA Meng;ZHENG Lu;ZOU Jinlin(College of Computer Science&Hubei Provincial Engineering Research Center for Intelligent Management of Manufacturing Enterprises,South-Central Minzu University,Wuhan 430074,China)

机构地区：[1]中南民族大学计算机科学学院&湖北省制造企业智能管理工程技术研究中心,武汉430074

出　　处：《中南民族大学学报（自然科学版）》2022年第6期712-719,共8页Journal of South-Central University for Nationalities：Natural Science Edition

基　　金：湖北省技术创新专项重大项目(2019ABA101);武汉市科技计划应用基础前沿项目(2020020601012267)。

摘　　要：针对葡萄叶片病害检测漏检率高,检测效果不佳的问题,提出了一种基于YOLOv4模型改进的葡萄叶片病害识别算法YOLOv4-PSA-CA.改进算法引入PSA(Pyramid Split Attention)模块取代YOLO4网络中原有的3×3的卷积,实现多尺度特征提取;将CA(Coordinate Attention)模块嵌入颈部网络中,获取更丰富的跨通道信息和位置信息.为了验证改进算法的有效性,选取葡萄叶片常见的4种病害作为检测对象制作数据集,改进的YOLOv4算法在此数据集上平均准确率均值(mAP)达到84.07%,比原YOLOv4算法mAP提升了4.04%.实验结果表明,改进算法能够在实验环境和自然环境下对葡萄叶片病害进行有效检测,为葡萄病害及时精准防控提供了依据.Aiming at the problem of high missed rate and poor detection effect of grape leaf disease detection,an improved grape leaf disease recognition algorithm YOLOv4-PSA-CA based on YOLOv4 model was proposed.Pyramid Split Attention(PSA)module is introduced to replace 3×3 convolution in YOLO4 network to extract multi-scale features.Embed CA(Coordinate Attention)module in neck network to obtain richer cross-channel information and location information.In order to verify the effectiveness of the improved algorithm,four common grape leaf diseases were selected as detection objects to make data sets,and the improved YOLOv4 algorithm achieved 84.07%average accuracy(mAP)on this data set,4.04%higher than the original YOLOv4 algorithm mAP.The experimental results show that the improved algorithm can effectively detect grape leaf diseases in the experimental environment and natural environment,providing a basis for timely and accurate control of grape diseases.

关 键 词：YOLOv4模型 葡萄叶片病害 注意力机制 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]