Study on the Detection of Xixia Text Based on SC-DBNet Modeling

Text detection is an important link in the digitization of Xixia ancient books. However, the structure of Xixia text is complex, with many strokes and high similarity, at the same time, the ancient books have foxing spot, fading and other problems leading to the difficulty of its detection and identification. Because of the fuzzy background and irregular layout of the text area in the Xixia ancient books, the Existing detection methods are not high enough in the detection accuracy, other indicators, leakage detection and other difficulties. Therefore, we propose a Xixia text detection method based on SC-DBNet, introducing the Shuffle Attention (SA) mechanism to enhance the feature extraction ability, and using the Channel Ehancement Feature Pyramid Network (CE-FPN) module to strengthen the feature fusion capability, mitigate information loss and optimise the integrated features. Compared with other methods, the experimental results show that our method is more improved and effective, whether from the model size, F1, Recall or Precision.