UTTSR: A Novel Non-Structured Text Table Recognition Model Powered by Deep Learning Technology

To prevent the compilation of documents, many table documents are formatted with non-editable and non-structured texts such as PDFs or images. Quickly recognizing contents tables is still a challenge due to factors irregular formats, uneven text quality, complex diverse content. This article proposes UTTSR recognition model, which consists four parts: region detection, line detection recognition, sequence recognition. For Cascade Faster RCNN ResNeXt105 network implemented, using TPS (Thin Plate Spline) transformation affine correct image improve accuracy. DBNET used Do-Conv in FPN (Feature Pyramid Networks) speed up training. Text lines recognized CRNN without CTC module, enhancing performance. Table based on transformer combined post-processing algorithms that fuse structure sequences unit grid Experimental results show model outperforms compared methods. upgraded significantly improves accuracy previous state-of-the-art F1 score tables, reaching 97.8%.