基于改进YOLOv8n的黄瓜叶片病害识别

打开文本图片集

中图分类号：S126；S436.412 文献标识码：A 文章编号：2095-5553（2026）02-0094-0

Abstract：Inorder todetectcucumberdisease targets incomplex natural environmentquicklyandacurately，imingatthe problemoflowacuracyofcucumber leaf imagerecognitioncaused bydiferent weather，angle，directionand distance，this study constructed 662O images of cucumber leaf data sets，including downy mildew，powdery mildew，late blight and normal leaves，andproposed AKGAM—YOLOv8 model basedon YOLOv8n model.Firstly，the variable kernel convolution AKConv is introduced into the Botleneck network layer toreduce the model parametersandcomputational overhead，making theoriginal modelmore lightweight.On this basis，the feature fusion network BiFPNand the attntion mechanism GAMareused toimprove themodel'sabilitytoextractsmallfeatures without reducing thedetectionspeed. Bychanging the lossfunctionof theoriginal modeltothe WIOUloss function，thegradient descent speedandthe loss valueafter convergenceare superior totheoriginal model.Theexperimentalresultsshow thattheacuracyrateof cucumber disease recognition of the improved model is 97.21% ，and the model weight is 13. 22 MB. Compared with the original model，the model weight is reduced to 56.42% of the baseline network，the accuracy rate is increased by 1.71% ， and the average accuracy is increased by 3.04% . It meets the requirements of real-time detection of cucumber diseases， and provides a theoretical basis for crop disease recognition and detection in complex natural environments.

Keywords：cucumber leafdisease；targetdetection；YOLOv8n；lightweight；AKConvconvolution；atention mechanism

0 引言

业生产中具有重要地位[1.2]。（剩余11278字）