燕山大学学报2025，Vol.49Issue(6)：487-495,9.DOI:10.3969/j.issn.1007-791X.2025.06.003
基于自适应原型解耦网络的小样本图像语义分割
Few-shot image semantic segmentation based on adaptive prototype decoupling network
摘要
Abstract
Aiming at the problem that the support and query images cannot make good use of in existing few-shot image semantic segmentation methods,an adaptive prototype decoupling network for few-shot image semantic segmentation is proposed.First,the backbone network with shared weights is used to map both the support image and the query image into a deep feature space,and the ground truth mask of the support image is utilized to decompose the support feature map into foreground and background features.Then,prototypes are generated in the support foreground feature map,and it is used to guide the segmentation of the unseen object in the query image.By comparing the contrastive loss between the predicted mask and the ground truth mask of the support image,the original prototypes are decoupled into main prototypes and auxiliary prototypes.Finally,an information interaction module is devised to establish the information interaction between the main prototypes of the support image and the global prototypes of the query branch,promoting the prototype alignment between support and query branches.Experiments conducted on the PASCAL-5i and COCO-20i datasets under 1-way 1-shot and 1-way 5-shot settings have validated the superior performance of the proposed method.关键词
图像语义分割/原型解耦/特征对齐/信息交互Key words
image semantic segmentation/prototype decoupling/feature alignment/information interaction分类
信息技术与安全科学引用本文复制引用
付瑞玲,李姿景,陈高威,钟德星..基于自适应原型解耦网络的小样本图像语义分割[J].燕山大学学报,2025,49(6):487-495,9.基金项目
国家自然科学基金资助项目(62376211) （62376211）
河南省高等学校青年骨干教师培养计划资助项目(2023GGJS185) （2023GGJS185）
河南省研究生教育改革与质量提升工程项目(YJS2023JD67) （YJS2023JD67）
