Open AccessA Multi-Criteria Band Selection Method of Hyperspectral Images
Author(s) -
Zhonghua Wang,
Yulin Zhao,
Bangsheng He,
Hao Wang
Publication year - 2020
Publication title -
journal of physics. conference series
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.21
H-Index - 85
eISSN - 1742-6596
pISSN - 1742-6588
DOI - 10.1088/1742-6596/1651/1/012147
Subject(s) - hyperspectral imaging , dimensionality reduction , geodetic datum , subspace topology , redundancy (engineering) , curse of dimensionality , pattern recognition (psychology) , artificial intelligence , computer science , selection (genetic algorithm) , dimension (graph theory) , mathematics , geography , cartography , pure mathematics , operating system
Hyperspectral remote sensing images are characterized by many bands and large amounts of datum, which require the dimensionality reduction, but band selection is one of the basic methods for dimensionality reduction of hyperspectral datum. For this reason, this paper proposes a multi-criteria band selection method. Firstly, the intrinsic dimensionality of hyperspectral image is calculated by virtual dimension, then the subspace is divided according to the band correlation criterion; and secondly, the information criterion is adopted to select the high-quality exponential bands in each subspace; and finally, the most suitable bands in each subspace are selected by class separability criterion to form the optimal band subset. Various experiments show that, compared with the optimum index factor and adaptive band selection methods, the proposed method has better performance in the measurement of information volume and information redundancy.