复杂环境下改进YOLOX的设施黄瓜病害检测方法

打开文本图片集

中图分类号：S436.421；TP391.41 文献标识码：A 文章编号：2095-5553（2025）08-0112-10

Abstract：Inresponse to the lowdetection acuracyand high miss rateofcucumberdiseases caused by factors suchas leaf blockage and overlap incomplex background environments，a cucumber disease detectionalgorithm called FSA EMAFPN—YOLOX is proposed.The FasterNet Block module is introduced in the feature extraction network，anda dual-branch structure attentionmechanism is embedded to suppressbackground noise，efectively solving theproblemof featureinformation losscaused by leaf blockage and overlap，andreducing the missrate.Inthe feature fusionstage，the EMA—AFPN feature fusion module is designed to reduce the lossof disease feature information.The SIoU bounding boxregression loss function is used toredefine the angle penaltymetric，which improves the training speedand bounding boxpredictionaccuracyofthemodel.TheVariFocalLossis introducedtosolvetheproblemof imbalanceddistributionof positiveandnegativesamples，enhancing themodel'slearmingof positive targetobjects，anditsfocusonthedisease area. Theresults showed that comparedwith the original YOLOX algorithm，the average acuracy of the improved YOLOX algorithm increased by 4.89% and the recall rate increased by 6.53% ，which significantly improved the detection effect of cucumber leaf disease under complex background. Keywords：facility cucumber;disease detection；attention mechanism；YOLOX；SIoU loss

0 引言

植物病害严重影响农业产量和质量，成为全球农业发展的焦点问题。（剩余14892字）