Two-Level Band Selection Framework for Hyperspectral Image Classification
- PDF / 1,530,873 Bytes
- 14 Pages / 595.276 x 790.866 pts Page_size
- 77 Downloads / 211 Views
RESEARCH ARTICLE
Two-Level Band Selection Framework for Hyperspectral Image Classification Munmun Baisantry1
•
Anil Kumar Sao2 • Dericks Praise Shukla2
Received: 11 June 2020 / Accepted: 8 November 2020 Ó Indian Society of Remote Sensing 2020
Abstract Dimensionality reduction strategies can be broadly categorized as band selection and feature extraction. Researchers and analysts from the remote sensing community give greater preference to band selection over feature extraction as the latter modifies the original reflectance values of hyperspectral data, making it difficult to understand the behavior of the materials in terms of their reflectance values. However, feature extraction strategies have their own advantages which cannot be ignored. Thus, a two-level, PCA-based band selection framework is proposed to unify the two dimensionality reduction strategies so that benefits of both the strategies can be derived. The proposed approach selects bands based on their relationship with a given set of principal components explained in terms of component loadings, thus keeping the original bands intact. Additionally, contrary to the popular notion that the complete information of all bands is adequately coalesced in the top principal components, middle principal components play a far stronger discriminative role when the competing classes are spectrally confusing to each other. Thus, for each level of classification, a different range of principal components is used to select the bands, on the basis of the level of spectral similarity expected between the classes at each level. Experimental results indicate that the proposed two-level band selection algorithm can select bands with varying levels of discriminative capabilities to effectively classify hyperspectral images consisting of classes spectrally very similar in nature. Keywords Hyperspectral Classification PCA Component loadings Band selection
Introduction Due to their high spectral resolution, hyperspectral imaging sensors have found tremendous usage in applications like mineral exploration (Cro´sta et al. 2000), forestry (Artigas and Yang 2005), geology (Ramakrishnan and Bharti 2015), defence (Cathcart 2008) and hydrological studies (Allegrini et al. 2005). However, an increased spectral resolution leads to significant data transmission, storage requirements and computational complexity to process the reflectances multitudinously. Further, many of these bands are & Munmun Baisantry [email protected]; [email protected] 1
Defence Terrain Research Laboratory, DRDO, Delhi, India
2
Indian Institute of Technology Mandi, Mandi, Himachal Pradhesh, India
contaminated by unwanted atmospheric influences and sensor-induced noise during the acquisition phase. Lastly, but most importantly, the number of samples required to estimate the parameters for modeling the information of hyperspectral images with confidence increases exponentially with the number of bands (Jia et al. 2013; Thenkabail et al. 2014). This phenomenon is termed as ‘Hughes
Data Loading...
