An Analysis Method for Recognizing and Analyzing Multi-type Documents in Electric Power Operation Site Based on PaddleOCR

The power operation site produces a large number of documents of various types and complex structures every day, and manual review of these documents is very time-consuming. In order to improve the review efficiency, a multi-type document recognition and analysis method based on PaddleOCR is proposed for the power operation site. First, the input documents are preprocessed, including format conversion and image correction. Then, the region of interest is obtained by PaddleOCR and classified into tabular and non-tabular types, which are recognized and analyzed using the tabular detection recognition model and traditional image processing methods, respectively. In particular, a multi-page document splicing method with text position adjustment by calculating page number information and average line spacing and a signature body recognition scheme based on feature extraction network and cosine similarity calculation are proposed to solve the challenges of multi-page document recognition and signature recognition. The experimental results verify the effectiveness of the method under different conditions, which significantly improves the efficiency of reviewing power operation documents.