摘要
为准确检测车辆尾灯灯语,实现模型轻量化,提出一种轻量化改进YOLOv5s网络模型。首先利用轻量级EfficientNet网络替换原主干网络,再将VoV-GSCSP模块替换颈部网络内C3模块,并在VoV-GSCSP模块后添加NAM注意力机制。针对上述3处改进,采用消融试验验证模型优化效果,模型训练采用车辆尾灯灯语检测专用的VLS数据集。实验结果表明,轻量化改进后模型相较于原YOLOv5s,参数量减少了41%、计算量减少了50%以及模型文件大小减少了39%,同时平均准确率精度(mAP@0.5)增加0.9%,证明改进后模型具有较好车辆尾灯灯语检测性能。
In order to accurately detect vehicle tail light signals and realize model lightweight,an improved lightweight YOLOv5s network model is proposed.Firstly,the lightweight EfficientNet network is used to replace the original backbone network,then the VoV-GSCSP module is used to replace the C3 module in the neck network,and the NAM attention mechanism is added after the VoV-GSCSP module.In view of the above three improvements,ablation test is used to verify the optimization effect of the model,and the VLS data set dedicated to vehicle tail light signal detection is used for model training.The experimental results show that compared with the original YOLOv5s,the number of parameters,calculation amount and model file size of the improved lightweight model are reduced by 41%,50%and 39%.Meanwhile,the average accuracy(mAP@0.5)is increased by 0.9%,which proves that the improved model has better tail light signal detection performance.
作者
祝磊
欧阳万棋
敖思铭
ZHU Lei;OUYANG Wanqi;AO Siming(College of Machinery and Automation,Wuhan University of Science and Technology,Wuhan 430081,China)
出处
《智能计算机与应用》
2025年第9期56-63,共8页
Intelligent Computer and Applications
基金
湖北省重点研发计划项目(2021BAA180)。
