There are some problems in traditional colorectal polyp image segmentation, such as large variation of lesion area scale, low contrast with normal tissue and insufficient ability to capture edge details. To solve the appeal problem, an adaptive polyp segmentation method based on phase sensing mixing is proposed. Firstly, Transformer encoder is used to reshape feature images, effectively establish feature spacing between short and long distance features, further extract detailed features and depth features, and output multi-scale resolution feature maps. Secondly, in order to enhance the ability to capture the location of the lesion area, the multi-scale feature information was cross-fused with clues to improve the correlation between global and local information, so that the network could effectively avoid the phenomenon of missing and missing in the segmentation of colorectal polyp pictures. Finally, the phase sensing hybrid module captures cross-level interaction information of each stage, effectively integrates global and local feature information, fills semantic gaps, suppositions background noise, and effectively aggregates multi-scale context information to complete polyp segmentation. The experimental results show that the new method can effectively segment colorectal polyp images, and has better performance in both subjective and objective evaluation.