基于YOLOv8的水稻虫害检测算法  

RicePest Detection Algorithm Based on YOLOv8

作　　者：刘贵锁 狄巨星[1] 杨阳[1] 王佳丽 LIU Guisuo;DI Juxing;YANG Yang;WANG Jiali(Hebei University of Architecture and Engineering,Zhangjiakou 075000,China)

机构地区：[1]河北建筑工程学院,河北张家口075000

出　　处：《长江信息通信》2024年第9期13-16,共4页Changjiang Information & Communications

基　　金：科研工作量评估与计算系统研发项目(No.2023CGIH01);基于YOLOv8的高速公路车辆跟踪算法研究(No.XY2024042);基于改进YOLOv8的水稻虫害检测算法研究(No.XY2024069)。

摘　　要：由于水稻虫害存在一些错检漏检的问题,文章提出了改进的YOLOv8模型BO-YOLOv8来提高水稻虫害识别的准确率。针对YOLOv8模型特征融合时浅层语义信息不足的问题,在Neck部分引入BiFPN网络,通过不同的权重来学习不同输入特征的重要性,提高水稻虫害特征表达能力。针对水稻虫害检测精度不理想的问题,在主干部分将原来的卷积替换为ODConv卷积,提高模型的检测精度。实验表明,BO-YOLOv8模型相较于改进之前mAP50提高了1.4%,精确度和召回率分别提升了2.0%和2.9%。实验结果表明,BO-YOLOv8模型可以有效提升水稻虫害的检测能力。As there are some problems of misdetection and omission of rice pests,the paper proposes an improved YOLOv8 model BO-YOLOv8 to improve the accuracy of ricc pest rccognition.Aiming at the problem of insufficicnt shallow semantic information during feature fusion in YOLOv8 model,BiFPN network is introduced in the Nck part to learn the importance of different input features through different weights to improve the feature expression ability of rice pest.For the problem of unsatisfactory rice pest detection accuracy,the original convolution is replaced with ODConv convolution in the Trunk part to improve the detection accuracy of the model.Experiments show that the BO-YOLOv8 model improves the mAP50 by 1.4% compared to the mAP50 before the improvement,and the precision and recall are improved by 2.0% and 2.9%,respectively.The experimental results show that the BO-YOLOv8 model can effectively improve the detection of rice pests.

关 键 词：YOLOv8 水稻虫害 目标检测 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]