基于改进YOLOv8模型的苹果品质检测算法

打开文本图片集

中图分类号：S451；TP391 文献标识码：A 文章编号：2095-5553（2026）04-0108-07

Abstract：Toensure theeficiencyandaccuracyof thetarget detection technology foroutdoorapplequality（whether infected with anthracnose or Marsonina blotch），a target detection model based on YOLOv8n—BSW was chosen to meetaseriesof requirements.Thismodel incorporatesa Bidirectional FeaturePyramid Network（BiFPN）module into the head network，which reduces the number of nodes in the network andenhances the feature fusion eficiency.In the backbone network of YOLOv8n，a Spatial Pyramid Pooling Enhanced by Local Atention Networks（SPPELAN） modulewasadded tosignificantlyimprovecomputational eficiencythrough segmentationandparalel procesingof apple surface features.The Wise—IoU（v2）lossfunction was also modified toimprove the accuracyand robustness of apple quality detection. The YOLOv8n—BSW model achieved a mean Average Precision （mAP）of 86.4% ， representing improvements of 2% ， 0.9% and 2.6% ，respectively，compared to the original YOLOv8n model. The YOLOv8n—BSW model shows significant enhancements in detection speedand computational efficiency，as well as excellent performance in detectionaccuracyand model structure optimization.The optimized model structure notonly improvestheeficiencyoffeature extractionandprocessing but also furtherreducescomputationalresource consumption，making it suitable for a wide range of practical applications.Theseenhancements enable the YOLOv8n一 BSW model to more rapidlyandaccurately identifyand clasify aples in orchards，providing strong technical support foreficient qualitycontrol.Moreover，themodel'sexcellent performancesignificantlycontributes tofoodsafety management，helping ensure the qualityof aple productsand providing a scientific foundation for orchard management and food safety assurance.

Keywords： apple;quality detection；deep learning；feature pyramid; loss function

0 引言

苹果品质检测在生产管理和市场流通中占据重要地位。（剩余13385字）