Medical image segmentation is one of the key technology of medical image intelligent analysis, where deep learning has contributed to medical image segmentation, based on deep learning proposed to improve Transformer medical image segmentation network and weight depth supervision (FS-TransUNet3 +), the model through the full scale jump connection structure, make the model aggregated multiple layers of abstract semantic features and spatial information, and reduce the network parameters of the model, improve the computing efficiency. At the same time, the weight depth supervision (Weight depth supervision-WDS) is adopted to improve the representation ability of the feature learning and the recognition accuracy of the image, refine the boundary of the target area, and improve the feature aggregation mechanism, splicing the semantic information of the hybrid encoder and the decoder, and strengthen the edge attention of the model in the image. The performance of each part is effectively validated on multiple datasets, such as bacterial image dataset, liver tumor segmentation challenge dataset, and multimodal brain tumor segmentation challenge dataset, and the comprehensive segmentation effect is better than other networks.