합성 데이터와 텍스트-심볼 통합 검출을 활용한 이미지 형식 P&ID 인식 기법

Abstract

A Piping and Instrumentation Diagram (P&ID) is a diagram used in the process plant industry. Digital format P&ID like intelligent P&ID can utilize DB technology, so it is easy to search and modify. Therefore, its use in the field has become common. However, there are cases in which digital P&IDs do not exist but exist only in image format because they were created before the digital P&ID was universalized or for security reasons. Thus, a technique for converting image format P&ID to digital P&ID is required. In this study, by modifying the deep learning-based symbol and text recognition structure presented in previous studies for symbol and text recognition of image format P&ID we propose a new structure that can improve performance while reducing the amount of computation required for recognition. In addition, we propose a synthetic data generation method suitable for P&ID in order to improve symbol recognition performance through data augmentation of the P&ID dataset. An experiment was performed to confirm the symbol and text recognition performance through a total of 82 P&ID drawings, and it was confirmed that the symbol and text recognition performance was improved through the method proposed in this study.