基于改进YOLOv8的冬枣成熟度检测方法

打开文本图片集

中图分类号：S665.1；TP391.4 文献标识码：A 文章编号：2095-5553（2025）12-0179-07

Abstract：Accurate detectionandmaturity discriminationof winter jujubeareprerequisites foryield estimationandrobotic picking ofwinterjujube.Todetectunripe，semired，and fullyripe winterjujubes intheorchard，this paper proposes SG一 YOLOv8，analgorithm basedon improved YOLOv8 for detecting theripeness of winter jujube fruits.The improvements includethree mainaspects.First，a global attention mechanism GAMis introducedintothebackboneof theoriginal networktoenhancethecorelationbetweenchannelstodistinguish winterjujubewithdiferentripenessmoreeffectively. Second，anaditionalsmallobjectdetectionlayerisadded totheneckofthenetwork toenancethedetectioncapabilityfor denselyclusteredandsmal-sized winterjujubes by expandingthedetectionscale.Inadition，a Slim-neck structure based on GSConv is constructed to reduce the memory occupation and computational complexityof the model while maintaining high accuracy.The experimental results show that the proposed SG—YOLOv8 modelachieves a mean average precision （204号 （mAP） of 95.00% ，which isa 2.7 percentage point improvement over YOLOv8s，10.9 over YOLOv6s，7.3 over YOLOv7，5.8 over RT—DETR—X，59.0 over Faster R—CNN，and 3.9 over YOLOv5s.

Keywords： deep learning；winter jujube；object detection;maturity detection； YOLOv8；feature fusion

0 引言

冬枣是一种营养丰富、口感独特的新鲜水果，富含维生素C，且具有重要的经济价值。（剩余10486字）