Abstract:In order to improve the accuracy of license plate detection of coal mine vehicles, an improved detection model based on YOLOv5s was proposed. In the feature fusion stage, the weighted bidirectional feature pyramid network (BiFPN) is used to add learnable weights to each input feature, so as to learn the importance of different input features and strengthen feature fusion at different levels. The EIOU loss was used to replace the CIOU loss of YOLOv5s network model as the boundary frame loss function, and the aspect ratio influence factors of the predicted frame and the real frame were separated, and the length and width of the predicted frame and the real frame were calculated respectively to accelerate the convergence speed of the network. CBAM attention mechanism is integrated into feature extraction network to extract more relevant feature information of target. The experimental results show that compared with the previous YOLOv5s model, the improved network model has improved the average accuracy (mAP) by 1.04%, and the convergence speed is faster. The improved model can effectively improve the accuracy of vehicle license plate detection.