计算机与现代化Issue(4)：81-87,7.DOI:10.3969/j.issn.1006-2475.2026.04.011
基于多特征图注意力的喉部反流语义分割网络
Multi-feature Map Attention Network for Laryngopharyngeal Reflux Semantic Segmentation
摘要
Abstract
Laryngoscopic image segmentation is crucial for aiding physicians in the diagnosis and treatment of LaryngoPharyn-geal Reflux Disease(LPRD).However,this task is confronted with three primary challenges:ambiguous boundaries,misjudg-ments arising from co-occurring features,and the low discriminability of small target regions from the background.To mitigate these issues,this paper proposes a novel multi-feature map attention-based semantic segmentation network for LPRD.The net-work integrates three key modules:a contrastive feature enhancement module to alleviate boundary ambiguity,a homogeneous region correction module to reduce misjudgments caused by co-occurring features,and a multi-feature fusion module to over-come challenges in small target region feature extraction.Experimental results on three laryngoscopic datasets demonstrate that the proposed model outperforms ten state-of-the-art segmentation networks,achieving mean Intersection over Union(mIoU)scores of 0.883,0.781,and 0.759 for granuloma,vocal fold sulcus,and laryngeal mucus,respectively.This approach effec-tively addresses the critical difficulties in laryngoscopic image segmentation,thereby providing robust technical support for the computer-aided diagnosis of laryngopharyngeal reflux disease.关键词
图像分割/咽喉反流疾病/多特征图注意力/语义分割Key words
image segmentation/laryngopharyngeal reflux disease/multi-feature map attention/semantic segmentation分类
信息技术与安全科学引用本文复制引用
许瑶,林思超,郑宝志,姚瀚晨..基于多特征图注意力的喉部反流语义分割网络[J].计算机与现代化,2026,(4):81-87,7.基金项目
福建省自然科学基金资助项目(2021J01388) （2021J01388）
福建省卫健委科技计划项目(2022ZQNZD001) （2022ZQNZD001）
