오픈소스 OCR을 이용한 군수품 시험성적서 전산화 연구

Abstract

Purpose: This study aims to transform test reports in the defense industry into a structured database (DB) by leveraging open-source Optical Character Recognition (OCR) and following the DMADOV methodology for quality improvement. Methods: The research was conducted in two phases following the DMADOV procedure. First, a baseline system was developed using the open-source OCR engine Tesseract to create a text extraction program, with data structuring attempted via rule-based post-processing. Subsequently, to overcome the system's limitations, a multi-model pipeline, specifically PaddleOCR's PP-Structure, was applied to enhance structural recognition performance, including layout analysis and table recognition. The performance of both systems was comparatively verified through quantitative metrics and qualitative analysis. Results: The initial Tesseract-based model heavily relied on strict, rule-based post-processing to ultimately achieve a 100% data match rate, but this revealed the system's lack of scalability and flexibility. In contrast, the optimized system using the multi-model pipeline (PP-Structure) accurately recognized the document's structure and content without requiring separate, complex post-processing, demonstrating superior performance in both qualitative and quantitative aspects. Conclusion: This study clearly identified the limitations of a simple OCR engine and demonstrated that a multi-model pipeline is an effective alternative for the automated structuring of defense quality data. The findings provide a practical roadmap for system integration companies and their partners to build a big data-based quality information system. Furthermore, the study is significant in its proposal of data utilization strategies for the implementation of Defense Quality 4.0.