Spectral matching accuracy in processing hyperspectral data

We investigated the accuracy of several spectral metrics when used for matching in hyperspectral imagery. Spectral matching refers to measuring the similarity among spectra. Two spectra are similar when the spectra distance between them is small and dissimilar otherwise. Spectral matching is used in many areas such as target detection and spectra classification. At the heart of spectral matching lay the distance measure used and the threshold value differentiating between similar and dissimilar. Frequent choices for such measures are spectral angle (SA) and spectral information divergence (SID) since they provide a limited range for threshold values. In addition, two new measures spectral correlation angle (SGA) and spectral gradient angle (SGA) are developed. Next, we suggest an alternative measure developed by using a normalized Euclidean distance (NED). To test the accuracy of the measures we used target information extracted from HYDICE data, and assessment tools such as the relative spectral discriminatory probability and the relative spectral discriminatory entropy. Experimental results suggest that NED outperforms SA, SCA and SGA and is relatively equivalent to SID. Given the reduction in computation time, NED constitutes an attractive measure to be used in spectra matching.