摘要
针对复杂路况背景下交通标志检测任务存在辨识度低、漏检严重等问题,提出一种基于改进YOLOv5s的轻量级交通标志检测模型。首先,引入坐标注意力模块,增强重要特征关注度;其次,对损失函数进行改进,降低边框回归时的自由度,加速网络收敛;最后,在中国交通标志检测数据集上进行实验。结果表明,模型在保持原有YOLOv5s模型体量的情况下,mAP@0.5提高了2.7%,检测速度达到91 FPS,对各种交通场景变化具有更好的鲁棒性。
Aiming at the problems of low recognition and serious leakage in traffic sign detection tasks in the context of complex road conditions,a lightweight traffic sign detection model based on improved YOLOv5s is proposed.Firstly,the coordinate attention module is introduced to enhance the attention of important features.Secondly,the loss function is improved to reduce the degree of freedom during border regression and accelerate network convergence.Finally,experiments are conducted on the Chinese traffic sign detection dataset.The results indicate that while maintaining the original YOLOv5s model volume,model's mAP@0.5 improves by 2.7%,with a detection speed of 91FPS,and it has better robustness to various traffic scene changes.
作者
李翔宇
王倩影
LI Xiangyu;WANG Qianying(Hebei University of Economics and Business,Shijiazhuang 050061,China)
出处
《现代信息科技》
2023年第10期30-32,36,共4页
Modern Information Technology
