基于YOLOv5的实蝇粘虫板检测方法  

Detection method of fruit flies trapped on sticky boards based on YOLOv5

作　　者：陈锦宏 秦睿嫔 刘小侠 张松斗 李贞 Chen Jinhong;Qin Ruipin;Liu Xiaoxia;Zhang Songdou;Li Zhen(Department of Entomology,College of Plant Protection,China Agricultural University,Beijing 100193,China)

机构地区：[1]中国农业大学植物保护学院昆虫学系,北京100193

出　　处：《植物检疫》2025年第2期12-19,共8页Plant Quarantine

基　　金：国家重点研发计划项目(2021YFC2600400);现代农业产业技术体系(CARS-28)。

摘　　要：实蝇已知种类约有4500余种,其中250余种具有经济意义,其幼虫取食果实,可导致果实腐烂或落果,发生严重时经济损失可达100%。粘虫板是田间实蝇发生情况调查、监测的重要手段之一,但是传统的人工识别计数效率低,成本高。为解决这一问题,本研究采用迁移学习的思路,使用YOLOv5算法训练了针对粘虫板上橘小实蝇、南亚实蝇和瓜实蝇的实蝇目标检测系统,对有3种实蝇的粘虫板检测的整体准确率和召回率分别达到86.9%和77.5%,其中橘小实蝇为84.9%和82.5%,南亚实蝇为82.2%和81.8%,瓜实蝇为93.5%和70.2%,可以在一定程度上代替人眼快速识别粘虫板上的实蝇种类和数量,为实蝇田间发生量监测提供了准确高效的方法。Fruit flies have over 4500 documented species,among which more than 250 species hold economic importance.The larvae of these pest species feed on fruits,potentially causing fruit rot or fruit drop.Under severe infestation conditions,such damage may result in complete(100%)economic losses in agricultural productions.Sticky boards are one of the most important means for investigating,and monitoring the occurrence of fruit flies in the field,but the traditional manual identification and counting is inefficient and costly.To solve this problem,this study adopted the idea of transfer learning and used the YOLOv5 algorithm to train a target detection system for B.dorsalis,Z.tau and Z.Cucurbitae trapped on sticky boards.The overall accuracy and recall rate of the detection for the three fruit species reached 86.9%and 77.5%,respectively,with 84.9%and 82.5%for B.dorsalis,82.2%and 81.8%for Z.tau,and 93.5%and 70.2%for Z.cucurbitae.The algorithm system could replace the human eye to a certain extent for rapid identification and counting of the fruit flies on the sticky boards,and provide an accurate and efficient identification method for monitoring the occurrence of fruit flies in the field.

关 键 词：实蝇 深度学习 YOLOv5算法 粘虫板 监测 

分 类 号：S41[农业科学—植物保护]