Traffic Sign Detection Based on Multi-scale YOLOv5
A YOLOv5 algorithm based on multi-scale fusion is proposed to address the problems of low detection accuracy and high missed detection rate of small object traffic sign detection.Four effective feature layers are output after the backbone network for better fusion of multi-scale information,and an improved multi-scale fusion attention mechanism CBAM_U is added to the three feature layers output from the backbone network to improve the detection capability of the network,followed by the addition of the Fusion module in the process of Path Aggregation Network(PANet)down-sampling to promote the fine fusion of features under different perceptual fields,the Adaptively Spatial Feature Fusion(ASFF)module is added before YOLOHead to address the inconsistency of feature pyramid fusion and further enhance the expression ability of the network.The experimental results show that the mAP@0.5 is improved by 3.07％,the recall rate by 3.83％,the accuracy by 1.64％,and F1-Score by 2.66％over the original YOLOv5 network on CCTSDB dataset,and the improved YOLOv5 algorithm has better robustness in complex scenes compared with other detection algorithms
万方数据
维普
