To solve these problems, this study adopts a residual grouped convolution module (RGCM) based on U-net, to reduce model parameters and computational load and accelerate model convergence.