Character Recognition System Based on Deep Learning Networks

Abstract

Appropriate algorithms and embedded compact mobile cameras have become the mainstream of current vision applications. The purpose of this paper is to solve the problem of book serial number recognition based on computer vision. Through supervised learning of specific book serial numbers, the characters are collected and stored in the Open CV library in the form of pictures, and the results of the files are manually marked. It is trained to recognize. For the target character picture, the serial number recognition is completed through image segmentation, character segmentation, and splicing. The recognition of characters by the model trained by the convolutional neural network algorithm is the research direction of this paper. After testing, the recognition time for a unit based on the camera installed on the Jetson Nano B01 (4b) motherboard is 0.05 s, and the recognition accuracy rate is as high as 94%. In reality, it has certain research value and application prospect.