Light-weight Traffic Sign Detection Algorithm Based on Attention and Contrastive Learning
To improve the accuracy and speed of traffic sign detection,a light-weight traffic sign detection algorithm based on attention and contrastive learning is proposed to address the issues such as small sizes and large-scale changes of traffic signs etc.Firstly,the channel and space separation method is used to do successive convolutions in the backbone network where features are extracted,and the network for multi-level feature extraction is built to reduce the amount of computation.Secondly,an attention-based contextual feature pyramid network is used to obtain representative features of target and improve the accuracy of the model.Finally,Supervised Contrastive Loss(SCL)is used to boost feature discrimination of the model.The experimental results show that the traffic sign detection algorithm achieves the average precision of 95.8%,4.5%higher than that of YOLOX-Tiny.The detection speed reaches 79 frames per second,which meets the need of practical applications.
traffic sign detectionfeature pyramidattentioncontrastive learningYOLOX-Tiny model
万方数据
维普
