Polyp Segmentation Method Combining HarDNet and Reverse Attention

A U-shaped colon polyp image segmentation network combined with HarDNet and reserve attention is proposed with the aim of solving the problems in the diversity of shape, size, color, and texture of colon polyps, the similarity between polyps and the background, and the low contrast of colonoscopy images, which affects the segmentation effect. The proposed model is based on the U-shaped encoder-decoder structure. First, the encoder uses HarDNet68 as backbone network to extract features for improving the reasoning speed and computational efficiency. Second, the decoder uses three reverse attention modules for fusing and refining the boundary features. Finally, multi-scale information fusion is realized between encoder and decoder through a receptive field module to provide more detailed edge information for the decoder. The iterative interaction mechanism between the encoder and decoder can effectively correct conflicting regions in the prediction results, improving the segmentation accuracy. The experimental results show that compared with existing methods, the proposed method improves segmentation accuracy and also has good real-time and generalization ability. The research results can provide a reliable basis for the early screening of colonic polyps.