重庆工商大学学报（自然科学版）2026，Vol.43Issue(4)：28-34,7.DOI:10.16055/j.issn.1672-058X.2026.0004.004
融合注意力机制和多层次特征的结肠息肉分割方法
Segmentation of Colon Polyps by Integrating Attention Mechanisms and Multilevel Features
摘要
Abstract
Objective To address the issue that the diversity of polyp shapes and sizes in colon polyp images and the low contrast between polyps and their surrounding tissues affect the precise segmentation of polyps,a colon polyp image segmentation method based on the DeepLabv3+network model that integrates attention mechanisms and multilevel features is proposed.Methods Firstly,the BAM_ASPP module was proposed and the BAM attention mechanism was added to the atrous spatial pyramid pooling(ASPP)module,so as to obtain important semantic information and improve the feature representation ability of the model.Secondly,after the feature fusion of the atrous spatial pyramid pooling module,the shuffle attention(SA)mechanism was introduced to obtain the key features.Finally,the convolutional block attention module feature fusion(CBAMFF)module was proposed,which fused the features of the third and fourth layers of the backbone network.Meanwhile,the attention mechanism was introduced to focus on location information to improve the segmentation accuracy of the model.Results Experimental results demonstrated that the mIoU and mDice of the proposed algorithm reached 91.63％and 95.54％,respectively,on the Kvasir-SEG dataset.Conclusion The segmentation performance of the proposed algorithm is superior to that of other compared methods.The proposed algorithm enables precise segmentation of colon polyp images,thereby assisting doctors in accurately removing polyps.关键词
息肉图像分割/DeepLabv3+/注意力机制/空洞空间金字塔池化Key words
polyp image segmentation/DeepLabv3+/attention mechanism/atrous spatial pyramid pooling分类
医药卫生引用本文复制引用
董坤朋,陈辉..融合注意力机制和多层次特征的结肠息肉分割方法[J].重庆工商大学学报（自然科学版）,2026,43(4):28-34,7.基金项目
安徽省重点教学研究项目资助(2020JYXM0458). （2020JYXM0458）
