Text Recognition for Automated Test Execution in Interlocking: A Deep Learning Approach

In this paper, we present a deep learning character recognition algorithm based on multi-level segmentation. It can improve the accuracy of recognition of button characters in the interlocked upper computer interface, which is of great significance to the design of the automatic test execution in the interlocking system. We first analyze the characteristics of characters in the interlocked upper computer interface and combine various methods to segment characters at multiple levels. Then, we build a deep learning character recognition model based on CNN to accurately recognize characters. Finally, we applied our algorithm to recognize the characters of the interfaces in a real interlocked upper computer software and made a comparison with other algorithms. The obtained results show the feasibility and advantage of our approach.