Vehicle appearance characteristics detection based on improving YOLOV4

In order to improve the accuracy of vehicle appearance feature information collection on manual toll passages in expressways and urban roads, this paper proposes an improved algorithm based on CBAM-EfficientNetV2-YOLOv4 to detect the appearance features of running vehicles. Aiming at the problem of single vehicle appearance features in the existing public dataset, a dataset of multiple vehicle appearance features is constructed. The vehicle appearance feature extraction algorithm proposed in this paper replaces the original feature extraction network CSPDarKNet53 with the backbone network of CBAM-EfficientNetV2-M, and replaces SPPNet with ASPPNet, so as to increase the receptive field of the model and extract feature information of different scales. The experimental results show that the mAP of the improved algorithm is 95.44%, which is 2.49% higher than the original algorithm, which shows that the algorithm proposed in this paper can effectively detect the appearance features of vehicles, and the accuracy rate is improved compared with other algorithms.