基于改进YOLOv5n的辣椒病害检测模型研究

打开文本图片集

中图分类号：S641.3；S436.418.1；TP391.4 文献标识码：A 文章编号：2095-5553（2025）10-0218-07

Abstract：Inorder torealize accurateand eficient identificationof pepperdisease，a pepper disease detection model based on improvedYOLOv5n is proposed.First，thebackbone network is replaced by theFasterNet network，which notonly reducestheredundantcalculation through partialconvolutionalPConv，butalsocaneffectivelyextractfeatures toenhance the model's abilityto expressimportant features.Secondly，combining the ELAN C2f module and Res2Net structure，the C2f-Res2 module is proposed tocarry out multi-scale featurefusion onthe input image at a deeper granularity level，and furtherstrengthen thefeature extractioncapabilityofthenetwork．Finally，WIoUloss functionisused toreplacethe originallossfunction CIoUto improve thequalityof anchor frame.The test results showed that the average accuracyof the improved YOLOv5n pepper disease detection model was 89.2% and the detection speed was 94.2 frames/s. Compared with YOLOv5n model，its mAP is increased by 5% ，accuracy rate and recall rate are increased by 4% and 5.9% respectively.Thismodel canefectivelyreduce theerrordetectionandleakagedetectionof pepper disease detection，and can identify and locate pepper disease more efciently.

Keywords：pepper disease;object detection； image processing；YOLOv5n

0 引言

以作为日常烹饪过程中的调味品，又可以作为药材使用，具有较高的经济价值1。（剩余10926字）