Improved U-net fundus image segmentation method

Abstract

The acquisition of medical images is difficult, and the small amount of data is a huge problem for image analysis. The uniqueness of U-net achieves good results on small samples. In this paper, U-net is used to segment vessels in the fundus image to predict some eye diseases early. The proposed U-net is changed to a seven-layer network from the classic one, and some parameters, such as patch size, are also optimized. The experimental results show that the fundus vessels obtained by this segmentation are very close to the marks, and the precision is better than other methods. The method has great significance for solving the segmentation problem of insufficient medical image data.