安徽农业科学2025，Vol.53Issue(3)：236-242,7.DOI:10.3969/j.issn.0517-6611.2025.03.047
一种基于改进YOLO v5n的黄桃虫害检测方法
An Improved Peach Pest Detection Method Based on YOLO v5n
摘要
Abstract
To reduce pest damage and increase peach tree yield,we proposed an improved peach pest detection method based on YOLO v5n.Firstly,a multi-class peach pest dataset was constructed to address the problem of overlapping pest generations and the long-tail distribution of pest data.Secondly,RFB(receptive field block)multi-branch dilated convolution layers were introduced into the backbone network to enlarge the model's receptive field and enhance its feature extraction capability.Simultaneously,the neck network structure was replaced with BiFPN(bi-directional feature pyramid network)structure to recombine features from different levels of feature maps and improve the way features were fused.Lastly,the activation function was changed to ReLU to avoid compatibility issues with certain hardware environments.Experimen-tal results showed that compared to the original YOLO v5n algorithm,the improved algorithm achieved a 1.6%improvement in accuracy,a 6.6%improvement in recall rate,and an average precision mean increase of 3.6%(reaching 88.8%).It outperformed other lightweight YO-LO models.关键词
桃/虫害检测/RFB/BiFPN/YOLO v5nKey words
Peach/Pest detection/RFB/BiFPN/YOLO v5n分类
农业科技引用本文复制引用
曾孟佳,徐欢,黄旭..一种基于改进YOLO v5n的黄桃虫害检测方法[J].安徽农业科学,2025,53(3):236-242,7.基金项目
教育部人文社会科学一般项目(20YJCZH005) （20YJCZH005）
浙江省湖州市工业攻关项目(2018GG29) （2018GG29）
湖州学院国家级大学生创新创业训练项目(202313287007). （202313287007）
