License Plate Detection Method Based on PLATE-YOLO in Smog Environment

Aiming at solving the problem of low license plate detection accuracy in the smog environment in the current license plate recognition field, a deep learning-based anti-smog license plate detection method is proposed in this paper. The method can detect civil license plates and airport civil aviation vehicle license plates. Firstly, a deconvolution algorithm based on convolutional neural network is used in the method to defog the license plate image. Then, the processed smog-free image is sent to the PLATE-YOLO network to detect the position of the license plate. The PLATE-YOLO network is a network for license plate detection that has been modified for the YOLOv3 network. The PLATE-YOLO network is suitable for the license plate detection network. There are two improvement points, one, a method for determining the number of anchor boxes based on hierarchical clustering algorithm is proposed. The other, for the characteristics of large license plate targets, the multi-scale feature fusion of the network is optimized. The optimized PLATE-YOLO network is more suitable for license plate detection and improves the detection speed. The experiment results show that the PLATE-YOLO network detects the speed of the license plate by 5 FPS compared with the YOLOv3. In the smog environment, the PLATE-YOLO network after dehazing pretreatment is more accurate than the unlicensed license plate detection method 9.2%.