Medical image segmentation can separate relevant important features from medical images and provide reliable basis for clinical diagnosis. In medical images, some parts are embedded in each other, and there is a problem of blurred borders or margins. Because of the difference of skin lesions in size, location, texture and color, it has become an important challenge in image segmentation in medical field. The multi-modal attention mechanism proposed in this paper enhances the recognition ability of the network by incorporating the channel attention mechanism, enriches the feature information at different scales, improves the accuracy of segmentation, and realizes the accurate segmentation of skin injury areas in dermatological medical images. Compared with many methods, our method achieves excellent performance.