基于改进YOLOv5s的樱桃成熟度检测模型  

Cherry Ripeness Detection Model Based on Improved YOLOv5s

作　　者：李一凡 刘从军 LI Yifan;LIU Congjun(School of Computer Science,Jiangsu University of Science and Technology,Zhenjiang 210034,China)

机构地区：[1]江苏科技大学计算机学院,江苏镇江210034

出　　处：《软件工程》2025年第4期47-52,共6页Software Engineering

摘　　要：针对智能采摘领域中樱桃成熟度检测的挑战,文章选取了YOLOv5s模型作为基础,并进行了针对性的改进和优化,旨在提升模型在樱桃成熟度检测任务中的性能和准确性。首先采用GhostNet替代主干网络,其次增加了NAMAttention模块,显著减少了模型的参数数量和计算需求,并且提升了检测准确性。实验结果显示,改进后的模型与YOLOv5s相比,每秒帧数提升了34%,参数数量减少了25%,FLOPs降低了29%,并且在mAP@0.5指标上实现了约6.2%的提升。此外,与多阶段模型相比,该模型在平均精度、效率和计算负载方面均表现优越。由此可见,改进后的模型能够更加高效、精准地进行樱桃成熟度识别。In response to the challenges of cherry ripeness detection in the field of intelligent harvesting,this paper selects the YOLOv5s model as the foundation and introduces targeted improvements and optimizations to enhance the model's performance and accuracy in cherry ripeness detection tasks.Firstly,GhostNet replaces the original backbone network,followed by the integration of a NAMAttention module,significantly reducing the model's parameter count and computational demands while improving detection accuracy.Experimental results demonstrate that,compared to YOLOv5s,the improved model achieves a 34%increase in frames per second,a 25%reduction in parameter count,a 29%decrease in FLOPs,and an approximate 6.2%improvement in the mAP@0.5 metric.Furthermore,compared to multi-stage models,the proposed model exhibits superior performance in average precision,efficiency,and computational load.These findings indicate that the improved model can perform cherry ripeness recognition more efficiently and accurately.

关 键 词：目标检测 果实成熟度检测 YOLOv5s Ghostnet NAMAttention 

分 类 号：TP399[自动化与计算机技术—计算机应用技术]