License plate detection based on improved YOLOv5 object detection

License plate detection is an important task in intelligent transport systems, which has a wide range of applications in vehicle management and public safety. Aiming at the current problems of low accuracy and efficiency and poor generalization ability of number plate detection whether based on traditional image processing methods or deep learning methods, we propose a number plate detection model based on improved YOLOv5 target detection. Firstly, MobileViTv3 is used to replace the backbone network of YOLOv5, which makes the model lighter while ensures the detection accuracy and allows the fusion of more features, then WIoU and AconC are used to further improve the accuracy of the target detection, and finally the Biformer bidirectional attention mechanism is introduced, which can help the model to better capture the correlation between the targets and the regional features, thus improving the detection of small and occluded targets. The experimental results of the article under the CCPD dataset show that the improved YOLOv 5 network achieves a detection accuracy of $\mathbf{9 9. 8 \%}$, a recall rate of $\mathbf{9 9. 9 \%}$, and an average precision of $99.5 \%$, which is significantly better than other number plate detection methods.