To solve the problems of false recognition of colon polyp region,blurred boundary location and difficult segmentation of complex samples,this paper proposes a colon polyp segmentation method based on dual-aggregation and agent attention.In this method,the context information is extracted by agent pyramid vision transformer (Agent-PVT),and the agent attention is used to reduce the amount of calculation while retaining the global modeling capability.A global-to-local aggregation module (GLAM) is proposed to capture both global and local texture features of the feature map.At the same time,the boundary aggregation module (BAM) is introduced to efficiently aggregate boundary and semantic information.In addition,batch nuclear-norm maximization (BNM) is introduced into the loss function to enhance the discrimination ability of complex samples.Experimental results on five datasets show that the proposed method has good performance in polyp segmentation.On Kvasir-SEG datasets,mDice and mIoU reach 92.83% and 88.16% respectively.