Small Vehicle Recognition Based on Attention Mechanism and Feature Fusion

This paper proposes an improved method combining the attention mechanism and feature fusion on the basis of YOLOv5. We firstly introduce CA attention mechanism in the network, which simultaneously considers the attention in channel dimension and spatial dimension; feature fusion adopts ASFF fusion mechanism, which solves the inconsistency problem inside the feature pyramid by learning the connection between different feature maps; finally, Focal Loss function is introduced during the training time. Based on the improved YOLOv5, the accuracy in identifying small vehicle targets is improved by 4.92%, and the processing speed of a single image is improved by 30.34%, which effectively enhances the processing accuracy and speed of small vehicle targets.