Application of U-Net with Filter in Rectal Tumor Recognition

U-NET is a full convolution network composed of convolution and upsampling. It is widely used in medical image recognition and segmentation. The basic idea of U-NET is to transform image segmentation into classification. The core problem of U-NET is still to solve the optimal parameters of loss function so as to facilitate the subsequent reverse propagation steps. But when the loss function is not convex, the problem often becomes very complex. The main purpose of this paper is to use U-NET neural network model with filter to solve the problem of accurate recognition and segmentation of tumor regions. By introducing filter theory, some theoretical defects of traditional algorithm can be avoided, and the problem of effective division of rectal tumor region can be solved.