基于改进YOLOv5s的咖啡果实成熟度检测方法

打开文本图片集

中图分类号：S571.2；TP391.4 文献标识码：A 文章编号：2095-5553（2025）10-0224-06

Abstract： Toaddress thechallenges of lowaccuracyand highcomputational complexity incurrent cofee fruit maturity detection methods，acoffeefruit maturitydetectionalgorithm based onan improved YOLOv5s model wasproposed.First，the YOLOv5s model was integratedwiththeShufleNet V2lightweight network toreducecomputationalcomplexity.Second，the SPPF module was embedded into the backbone network to enhance modelscapabilityfor feature information extraction and fusion.Additionally，the RFAConv atention convolution was incorporated toimprove the model's focuson spatial features within thereceptive field，further enhancingdetectionperformance.Finally，tooptimize learmingoftargetpositionand shape information，the CIoU loss functionwas replacedwith the SIoU loss function，which contributed to improved detection accuracy of the model. The experimental results showed that the improved YOLOv5s algorithm achieved a 0.5% increase in precision，a 0.1% increase in recall，and a 1.1% improvement in mean average precision （mAP ）compared to the original model.Furthermore，thealgorithm'sFLOPswerereduced by 10% ，and its computational complexity decreased by 4.4% This optimization not only enhanced the model's detection accuracy，but also reduced algorithmic complexity.

Keywords：coffee fruit；maturity detection;deep learning；YOLOv5s；target detection;attention mechanism

0 引言

国家和地区种植咖啡[1]。（剩余10239字）