Multisize plate detection algorithm based on improved Mask RCNN

Abstract

The sorting of plates is an indispensable part of the plate processing production line. In order to achieve plate detection in complex detection scenarios, a multisize plate detection algorithm based on improved Mask RCNN is proposed. The model fusion method is used to introduce the DenseNet network structure to optimize the feature transfer path to make feature extraction more efficient. At the same time, the boundary distance constraint is added to the segmentation loss function, which makes the model more precise for the target with high stacking complexity and fuzzy boundary information. The experimental results show that the improved Mask RCNN performance is significantly improved, compared with other models, it achieves an optimization effect with an average accuracy of more than 98%.