Objective To improve the segmentation performance of the intestinal polyp images semantic segmentation model for unknown objects in the query images.Methods A semantic segmentation method for few-shot intestinal polyp images based on cross-cooperative attention network was proposed.Firstly,the pre-trained VGG-16 network was used to extract the visual features of supporting and query images.Then,the cross-integration of inter-branch features was established by using support features and query features to promote the semantic alignment of inter-branch features.Finally,the pixel-by-pixel classification of each position in the query image was realized by using the non-parametric measurement method.Results In Kvasir-SEG and three other open-source intestinal polyp image datasets,the proposed method achieved superior FB-IoU scores compared to that of U-net,a classical semantic segmentation model for medical images.Conclusion The semantic segmentation method of few-shot intestinal polyp images based on the cross-cooperative attention network can accurately locate the polyp region in the supporting images and query images,demonstrating good segmentation performance.
维普
万方数据
