Vehicle Target Detection Algorithm Based on CBAM-YOLOv5s

Aiming at the concerns regarding low accuracy and lack of real-time performance of vehicle recognition in the field of automatic driving in the case of complex roads and occlusion between targets, this study suggests an advanced YOLOv5s network model. Firstly, multi-scale data augmentation was performed on the input layer to expand the size of the training data set and enrich the features of the image. Secondly, the Loss function of the network was optimized, and the SIoU function had become proficient in replacing the CIoU Loss function of YOLOv5s original network. Finally, the CBAM attention mechanism was introduced into the feature extraction network, and the redundant channels of the network were pruned to improve the detection accuracy and detection speed of the model. This paper presents an improved YOLOv5-s model, and experimental results demonstrate that it significantly enhances the accuracy of vehicle target detection while ensuring the detection speed. On the selected data set in comparison to the original YOLOv5s, the suggested network achieved a 5.1 percentage point improvement in mAP0.5, a 3.4 percentage point improvement in Precision, and a 4.8 percentage point improvement in recall value. Consequently, the paper’s enhanced YOLOv5-s network successfully completes the vehicle target recognition task in the autonomous driving domain.