文献类型：期刊文献

英文题名：Fusion of global and adaptive local information for few-shot image classification

作者：Xiao, Ting[1,2];Xia, Yiqing[2];Tang, Ruiqi[2];Du, Wenli[1,2];Wang, Zhe[1,2]

机构：[1]Minist Educ, Key Lab Smart Mfg Energy Chem Proc, Shanghai 200237, Peoples R China;[2]East China Univ Sci & Technol, Dept Comp Sci & Engn, Shanghai 200237, Peoples R China

年份：2025

卷号：168

外文期刊名：PATTERN RECOGNITION

收录：;EI(收录号：20252118480074);WOS:【SCI-EXPANDED(收录号:WOS:001504586500004)】；

基金：This work is supported by the National Natural Science Foundation of China under grants No. 62476087 and No. 62306115, and the National Key Research and Development Program of China under grant No. 2022YFB3203500.

语种：英文

外文关键词：Few-shot learning; Image classification; Fusion of global and local feature; Knowledge collaboration

摘要：Meta-learning-based few-shot learning methods enable to quickly and accurately adapt to new tasks by simulating diverse scenarios and leveraging prior experiences. However, they face challenges in optimization and handling intra-class variability. Additionally, it often overlooks the interaction between global and local information. This paper proposes a Fusion of Global and Adaptive Local Information (FGAL) method to address challenges in model optimization and intra-class variations by fully considering the beneficial impacts and interactions between global and local views. FGAL includes two key modules: the classifier re-weighting optimizer and the global-local knowledge collaboration module. The former improves model optimization by refining training strategies and using a multi-head attention mechanism with a broader receptive field to adaptively process query images. The latter calibrates the training bias between global and local networks, facilitating synergistic knowledge optimization and significantly improving generalization. Experimental results on three few-shot datasets and a cross-domain dataset validate our approach's effectiveness and adaptability.