Band Subset Selection for Hyperspectral Image Classification

Abstract

This paper develops a new approach to band subset selection (BSS) for hyperspectral image classification (HSIC) which selects multiple bands simultaneously as a band subset, referred to as simultaneous multiple band selection (SMMBS), rather than one band at a time sequentially, referred to as sequential multiple band selection (SQMBS), as most traditional band selection methods do. In doing so, a criterion is particularly developed for BSS that can be used for HSIC. It is a linearly constrained minimum variance (LCMV) derived from adaptive beamforming in array signal processing which can be used to model misclassification errors as the minimum variance. To avoid an exhaustive search for all possible band subsets, two numerical algorithms, referred to as sequential (SQ) and successive (SC) algorithms are also developed for LCMV-based SMMBS, called SQ LCMV-BSS and SC LCMV-BSS. Experimental results demonstrate that LCMV-based BSS has advantages over SQMBS.

Unsupervised hyperspectral band selection in the compressive sensing domain

“…' ∈ £ % ∈ ¡ (10) The similarity measure between all bands is depicted in figure 3 rendering all values of ( o , : ) when each band is a separate cluster. These matrices suggest the clustering which exists due to the highly correlated spectral bands.…”

“…The key insight to CSBC is the preservation of the ED and SAM using the RIP and RCP properties of CS used to compute the distance between clusters in (10). An experiment was conducted to show the accuracy by computing the difference between the similarity matrices using the ODS and the sensed bands.…”

“…This error metric is an unnormalized Tanimoto index and used to estimate the accuracy of the SBS when using CS and without [43]. Table 1 shows the SBS using the Euclidean distance between clusters in (10). When the CSSR is inadequate, the CBR is lower and the SBS is erroneous.…”

“…Band selection (BS) has an advantage over other DR algorithms because it does not change the semantics of the image samples which allows for HSI data processing directly on the reduced data set [9]. In addition, many exploitation algorithms have minimal performance loss when applied to the selected band subset (SBS) [10], [11]. However, BS is a feature selection optimization algorithm and can be computationally intensive.…”

“…Orthogonal subspace projection as well as mutual information have been explored to solve BS using band decorrelation [23], [24]. Finally, a plethora of search strategies have been put forth in the literature [10], [25], [26].…”