基于子空间划分和自我表示学习的高光谱波段选择  

Hyperspectral band selection based on subspace partition and self-representation learning

作　　者：王鑫 汪国强[1] WANG Xin;WANG Guoqiang(College of Electronic Engineering,Heilongjiang University,Harbin 150080,China)

机构地区：[1]黑龙江大学电子工程学院,哈尔滨150080

出　　处：《黑龙江大学自然科学学报》2021年第2期228-237,共10页Journal of Natural Science of Heilongjiang University

基　　金：国家自然科学基金资助项目(51607059);黑龙江省自然科学基金资助项目(QC2017059)。

摘　　要：基于高光谱图像的应用,在不降低性能的前提下,选择具有信息和代表性的波段是大数据环境下一项具有挑战性的任务。许多波段选择方法忽略波段的有序性,只考虑波段的冗余性,这会导致有价值的波段丢失,而保留无用的波段。针对此问题,提出了一种自适应子空间划分和自我表示学习的高光谱波段选择方法(Adaptive subspace partition-Self-representcd learning,ASP-SRL),该方法最大限度地利用类间距离与类内距离之比,将高光谱图像立方体分割为多个子立方体。子立方体采用自表示学习算法处理,在处理完所有子立方体以后,采用记忆向量q进行波段选择。与三个公开高光谱影像数据集和最新的波段选择方法Multi-dictijonary sparse representation(MDSR)、Scalable one-pass self-representation learning(SOP-SRL)、Adaptive subspace partition strategy(ASPS)相比,所提出方法在OA、AA和Kappa三个指标上都优于其他算法。For applications based on hyperspectral images,it is a challenging task to select the informational and representative bands in the context of big data without reducing performance.However,for many band selection methods,ignoring band order and considering only band redundancy will result in the loss of valuable bands and retaining useless bands.To solve these problems,an adaptive subspace partition and self-representation learning hyperspectral band selection method(Adaptive subspace partition-Self-representcd learning,ASP-SRL)is proposed.This method makes maximum use of the ratio of inter-class distance and intra-class distance to divide the hyperspectral image cube into multiple subcubes.Then the subcubes are processed by the self-representation learning algorithm.After all the subcubes are processed,the memory vector q is used for band selection.The effectiveness of the proposed method is verified by comparing it with the latest band selection methods Multi-dictijonary sparse representation(MDSR),Scalable one-pass self-representation learning(SOP-SRL),Adaptive subspace partition strategy(ASPS)on three public hyperspectral image datasets.The experimental results show that the proposed method is superior to other algorithms in OA,AA,and Kappa indices.

关 键 词：自适应子空间 波段选择 自表示学习 高光谱图像 

分 类 号：Q939.97[生物学—微生物学]