Research on image segmentation techniques based on algebraic topology methods in computer vision

Medical image segmentation is the basis for realizing intelligent medical treatment, and plays a very important clinical significance in the localization and identification of lesion areas and the formulation of surgical plans. In this paper, we investigate the image segmentation techniques based on algebraic topology methods in computer vision, and propose an image segmentation network model based on asymmetric topology preservation (ATSNet), with a view to applying it to clinical practice.The ATSNet model adopts the parallel branching structure of CNN and Transformer in the coding part, and proposes a hybrid feature aggregation strategy (HFAS) to achieve image segmentation with high efficiency. Comparison experiments on three benchmark datasets and one clinical dataset prove that the ATSNet model proposed in this paper achieves better results on different datasets, and the statistical analysis results obtained by the model are consistent with those of clinical experts (P>0.05). Meanwhile, ablation experiments demonstrate the effectiveness of the hybrid feature aggregation strategy used in this paper in improving the image segmentation performance of the model. In addition, the proposed method in the Transformer branch when the number of network layers is 3 when the overall accuracy of the largest, and the use of bilateral filtering can be better edge retention, improve the effect of image segmentation. This paper provides a technical path for the practical application of image segmentation technology.