Colon Polyp Segmentation Method Based on Short-Term Dense Concatenate Attention Network
Colonoscopy is operator-dependent and has a high missed detection rate,so a real-time polyp segmentation algo-rithm is needed to assist doctors in polyp screening.Therefore,this paper proposes a short-term dense concatenate attention net-work.The core layer of the network encoder is the short-term densely connected attention module.This module integrates traditional convolution,STDC Module,residual learning and NAM module,retains scalable receptive field and multi-scale information with small computational complexity.PD is introduced at the decoded path,some low-level features are discarded for model accelera-tion,and high-level features are aggregated to achieve better segmentation results.STDCANet compares the performance and model complexity with the classical medical image segmentation network on the CVC-ClinicDB dataset.It is superior to the comparison net-work in both aspects and has the potential for clinical real-time segmentation.
deep learningmedical image processingattention mechanismcolonoscopy images
万方数据
维普
