A single-stage automatic license plate recognition network with Balanced-IoU loss

Abstract

In this paper, we propose a license plate recognition model, which can detect and recognize the license plate in a single forward. The features of the input image are extracted by our 15-layer convolutional neural network. In detection branch, we use a loss function with better nonlinear to fit the detection process of license plate. To catch the location of license plate with less information loss, we add Intersection over Ground-truth (IoG) into the Intersection over Union (IoU) and get Balanced-IoU loss (BIoU loss). The combination of these two loss functions can make the model get a better predictive result. In recognition branch, we introduce an attention mechanism to make better recognition results. The experimental results show that our network can get 99.91% recognition precision, which is over 3% higher than state-of-the-art single-stage ALPR network RPnet.