基于改进YOLOv5s的轻量化草莓成熟状态检测算法

打开文本图片集

中图分类号：S668.4；TS24；TP391.4 文献标识码：A 文章编号：2095-5553（2026）02-0086-08

Abstract：Toachieve the accurate detectionof strawbery ripeness innatural environments，taking strawberries as the main research object，a lightweight strawberry ripening status detection algorithm，PBW—YOLOv5s，basedon the improved YOLOv5s model，was proposed.Firstly，Partial Convolution（Pconv）was integrated into the C3 module，and the efective function（IoU） was replacedwith a dynamic non-monotonic focusing mechanism（WIoU）to more acurately considertheoverlapdegree betweenthepredictedboxand therealbox，which furtherlyefectivelyreducedthenumberof network parameters andaddressed iregularocclusions.Secondly，the WeightedBi-directional Feature Pyramid Network （BiFPN）was introduced to enhance the integration offeatures acrossvarious scales.Lastly，thebounding box regression wasimprovedtoenhancethedetectionperformance.Experimentalresults showed thatthe precisionrate，recallrate，and mean Average Precision（ ;mAP ）ofPBW—YOLOv5s were increased by 2.49% ， 1.21% ，and 1.06% ，respectively. Concurrently，the parameter count and model size were respectively reduced to 80.06% and 80.00% of the original YOLOv5s.Additionaly，testsonthevalidationsetdemonstratedthatthis detectionalgorithm efectively prevented missed andfalsedetections while identifying strawberry ripeness moreaccurately，thus providing technical support for subsequent intelligent strawberry harvesting and management.

Keywords：strawberry ripeness；image recognition；machine vision；object detection；lightweight algorithm

0 引言

我国作为全球最大草莓生产国，近年来，随着草莓需求量持续增长，种植面积和产量亦相应扩大[1。（剩余14173字）