This paper proposes a mask segmentation and evaluation method based on Mask Scoring R-CNN, which combines Mask R-CNN and the mask evaluation mechanism. We adopt Mask R-CNN to extract accurate target masks from images, and propose a mask evaluation mechanism to evaluate the quality of the generated masks. By presenting mask scoring, we are able to accurately measure the accuracy and completeness of masks, thereby improving the quality of mask segmentation. In order to ensure the accuracy and completeness of segmentation, we have optimized Mask Scoring R-CNN by adding a loss function that can be used for multi-category image segmentation. This loss function can improve the segmentation task. The evaluation index IoU is optimized and designed to improve the original mask quality score while maintaining a high accuracy of the segmented images.