基于改进YOLOv8的香菇成熟度检测方法

打开文本图片集

中图分类号：S646.1十2 文献标识码：A 文章编号：2095-5553（2026）03-0074-07

Abstract：AccuratedetectionofLentinulaedodesmaturityusingmachinevisionisakeyprerequisiteforinteligent harvesting.However，the mushroom’s small size and complex growing environment create high demandsfor detectionaccuracy，particularlyin smal-targetrecognition.Toaddressthese chalenges，thispaper proposesa mushroom maturity detection method basedon an improved YOLOv8 instance segmentation model.First，the BiFormeratention mechanismis integratedinto thebackboneofYOLOv8 toenhance the model'sability toextractcitical feature.Second，theP2 feature map from the backbone network isintegrated intothe model's feature fusion network， enrichinglow-levelfeaturesandimprovingboththedetectionaccuracyandsmal-targetdetectionperformance.Finally， experiments on Lentinula edodes maturity detection were caried out.Theresults show that the improved YOLOv8 algorithm can accurately identify the maturity of Lentinula edodes，achieving a detection accuracy of 95.30% and an mAP@0.5 of 96.51% .Overall，the proposed method outperforms other existing vision-based detection models.

Keywords：Lentinula edodes；maturitydetection；smallobjectdetection；atention mechanism；backbone network

0 引言

香菇（Lentinulaedodes）作为一种可食用菌，因其营养丰富，食用价值高等优点，在全球范围受到广泛欢迎[1]。（剩余10303字）