摘要
针对车检站中车辆检测的实际需求,提出一种改进YOLOv5s的轻量级车脸检测算法。使用ShuffleNetV2网络作为Backbone,在保证模型检测精度的同时实现模型的轻量化与实时性;将通道-空间注意力(SA-Net)与跨通道注意力(Triplet)相结合,提出一种跨通道-空间注意力模块(SA-Triplet attention, STA),提高模型的检测精度;提出一种基于STA注意力模块的跨层特征融合模块(SA-Triplet attention feature fusion, STA-FF),进一步提高模型的检测精度。在自建车脸检测数据集Car-Data上进行实验,所提模型的平均检测精度达到了94.3%,检测速度达到了105.60 FPS,模型参数量为12.36 M。
Aiming at the actual demand of vehicle detection in vehicle inspection station,a lightweight car front detection algorithm was improved by YOLOv5s was proposed.ShuffleNetV2 network was used as Backbone to realize the lightweight and real-time of network model while ensuring the accuracy of model detection.A SA-Triplet attention module(STA)was proposed by combining SA-Net and Triplet Attention to improve the detection accuracy of the model.A cross-layer feature fusion module(STA-FF)based on STA attention module was proposed to further improve the detection accuracy of the model.Experiments were carried out on Car-Data,a self-built car front detection dataset,the average detection accuracy of the proposed model reaches 94.3%,the detection speed reaches 105.60 FPS,and the parameters of model are 12.36 M.
作者
余国豪
贾玮迪
余鹏飞
李海燕
李红松
YU Guo-hao;JIA Wei-di;YU Peng-fei;LI Hai-yan;LI Hong-song(School of Information Science and Engineering,Yunnan University,Kunming 650500,China)
出处
《计算机工程与设计》
北大核心
2024年第3期732-739,共8页
Computer Engineering and Design
基金
国家自然科学基金项目(62066046)。