Single-layer and double-layer license plate recognition method based on multi-task convolutional neural network

Abstract

The invention discloses a single-layer and double-layer license plate recognition method based on a multi-task convolutional neural network, and the method comprises the steps: automatically generating license plate data of each province, and carrying out the image processing of the generated license plate data through an image processing program; inputting the single-layer license plate picture and the double-layer license plate picture into a deep learning model, and processing through a convolutional neural network in the model to obtain feature maps F1, F2 and F3; performing convolution pooling operation on the feature map F1 to generate a feature map F4 of a single-layer license plate; performing convolution pooling operation on the feature map F2 to generate a judgment feature map C1, and performing feature classification on the judgment feature map C1; performing convolution pooling operation on the feature map F3 to generate a feature map F5 and a feature map F6; splicing the feature maps F5 and F6 to obtain F56; and carrying out weighted summation on the feature map F4 and the spliced feature F56 to obtain an extracted license plate feature F7. According to the invention, the whole architecture dynamically selects to use single-layer or double-layer features by using a multi-task model through a feature map, so that single-layer and multi-layer license plate recognition can be realized at the same time.

Description

technical field [0001] The invention relates to the field of license plate recognition, in particular to a single-layer and double-layer license plate recognition method based on a multi-task convolutional neural network. Background technique [0002] With the rapid development of social economy, motor vehicles are increasingly becoming an indispensable means of transportation for people's daily life. All provinces and inter-city expressways and main roads, city entrances and exits, and major traffic arteries have deployed bayonet equipment to collect information from passing vehicles. However, current bayonets are generally based on license plate recognition technology. Once a suspected vehicle uses fake License plates, sets of cards, no license plates, and constant replacement of license plates can evade the tracking and identification of existing bayonets. [0003] Image-based vehicle feature recognition involves image processing, pattern recognition, computer vision and...

AI Technical Summary

Problems solved by technology

[0007] 1. It is difficult to collect data. It is time-consuming and labor-intensive to collect license plate data from different provinces

[0008] 2. The implementation scheme based on deep learning convolutional neural network cannot realize the end-to-end simultaneous recognition of single-layer and double-layer license plates. It often requires three models to complete the license plate recognition, a model for judging single-layer and double-layer license plates, and a single-layer recognition model. model, a two-layer recognition model, and our current algorithm is an end-to-end model that uses multi-tasks to replace what the previous three models can accomplish.

[0009] Since the previous algorithm model needs three combinations to complete the license plate recognition, the execution speed of the algorithm will be greatly reduced, and the accuracy of multiple models in series will also decrease. For example, the accuracy of the single-layer and double-layer license plate model is 90%. The accuracy of layer license plate recognition is 90%, then the overall accuracy is 90%*90% = 81%

Images