计算机技术与发展2025，Vol.35Issue(11)：53-60,8.DOI:10.20165/j.cnki.ISSN1673-629X.2025.0170
基于方向感知与语义渐进优化的结肠息肉分割模型
Colon Polyp Segmentation Model Based on Direction-aware and Semantic Progressive Optimization
摘要
Abstract
Studies have shown that early detection of colon polyps can reduce colon cancer,and automated polyp detection is one of the key technologies.Due to the different sizes and shapes of intestinal polyps,coupled with the blurred demarcation between polyps and normal intestinal wall,which are more similar at the pixel level,it makes polyp segmentation challenging.To address the above problems,DCSPO-Net(colon polyp segmentation method integrating direction-aware context and semantic-aware progressive optimization)polyp segmentation model is proposed.Taking EfficientB5 as the backbone network for feature extraction,we design a multi-kernel adaptive encoding(MKAE)module to achieve synergistic enhancement of local details and global contextual information through dynamic weight allocation to enrich the information of polyp features,and a direction-aware context fusion(DACF)module to utilize a direction-aware mechanism to accurately capture the direction-sensitive features of the polyp morphology,and a semantic-aware progressive optimization Decoder(SAPRD)module fuses multi-scale-directional features by semantic attention weighting,and combines lightweight edge-enhanced convolution for asymptotic optimization of the features,which effectively improves the segmentation accuracy and boundary coherence.Experimental results show that the proposed model achieves better results on kvasir-seg,CVC-ColonDB,CVC-ClinicDB,CVC-300,and ETIS-LaribpolypDB datasets with better segmentation accuracy and boundary segmentation compared to the current mainstream segmentation models.关键词
息肉分割/DCSPO-Net/多尺度特征/方向感知/渐进优化Key words
polyp segmentation/DCSPO-Net/multi-scale features/directional-aware/progressive optimization分类
信息技术与安全科学引用本文复制引用
司静宇,辛梦楠,严恺,闫晓东,王瑞杰..基于方向感知与语义渐进优化的结肠息肉分割模型[J].计算机技术与发展,2025,35(11):53-60,8.基金项目
山西省基础研究计划项目(202103021223011) （202103021223011）
