Special Convolution and Mask R-CNN: A New Method for Text Recognition on Cultural Relics

This paper explores the use of artificial intelligence technology to restore and recognize blurred text on cultural relics. Traditional OCR technology has limitations in dealing with incomplete or blurred text. In contrast, this study proposes a new deep learning-based method that employs Feature Pyramid Networks (FPN), Region Proposal Networks (RPN), and Fully Convolutional Networks (FCN), along with a special convolution module, to effectively improve the accuracy of text recognition on cultural relics. The experimental part utilized datasets such as COCO and validated the method through the Mask R-CNN model. Evaluation metrics include classification loss, bounding box regression loss, and segmentation mask loss. The results show that this method has high robustness and accuracy in complex scenarios, providing a new solution for the protection and recognition of text on cultural relics.