利用多级知识蒸馏进行高光谱图像分类的终身学习方法  

A lifelong learning method for hyperspectral image classification using multi-level knowledge distillation

作　　者：姜紫惠 李照奎 王珂 JIANG Zihui;LI Zhaokui;WANG Ke(School of Computer Science,Shenyang Aerospace University,Shenyang 110136,China)

机构地区：[1]沈阳航空航天大学计算机学院,沈阳110136

出　　处：《激光杂志》2025年第3期98-104,共7页Laser Journal

基　　金：国家自然科学基金项目(No.62171295);辽宁省重点研发计划-应用基础研究项目(No.2023JH2/101300204)。

摘　　要：针对终身学习面临的灾难性遗忘问题,提出利用多级知识蒸馏进行高光谱图像分类的终身学习方法。首先,设计基于多模态对齐的特征提取器,充分利用高光谱图像的空谱信息和标签文本信息。此外,设计多级知识蒸馏策略,有效保留旧阶段的多模态知识。提出的方法在两个公共高光谱数据集上进行实验,与目前最新的方法相比,所提方法在Pavia University数据集上平均精度提升了15%~18%,在Botswana数据集上平均精度提升了1%~8%。To address the catastrophic forgetting problem in lifelong learning,this paper proposes a lifelong learning method for hyperspectral image classification using multi-level knowledge distillation.Initially,a feature extractor based on multi-modal alignment is designed to fully leverage the spatial-spectral information and label text information of hyperspectral images.Additionally,a multi-level knowledge distillation strategy is devised to effectively preserve the multi-modal knowledge from previous phases.The proposed method was experimented on two public hyperspectral datasets.Compared to current state-of-the-art methods,the proposed method showed an average accuracy improvement of 15%-18% on the Pavia University dataset and 1%-8% on the Botswana dataset.

关 键 词：高光谱图像分类 终身学习 多模态对齐 知识蒸馏 

分 类 号：TN911[电子电信—通信与信息系统]