A Study of Spectral Data Processing with Emphasis on Spectral Similarity Measures for Hyperspectral Image Processing

Abstract

In this paper, we have performed extensive research on the various methods which can be employed to develop the libraries for an automatic spectra verification software tool that is capable of classifying spectra in terms of their quality and correctness of the proposed parameter. A key step in building library is to establish the degree of similarity between the various bands of the spectrum, which may originate from various sensors or other data collecting devices. Various methods have been proposed for describing the spectral and structural similarity. Here, we have employed empirical mode decomposition (EMD) followed by spectral similarity measures based on amplitude and shape features. Non-stationary and non-linear series can be processed using the empirical mode decomposition. It is an adaptive time-space analysis method that employs the method of partition. Without leaving the time domain, intrinsic mode functions (IMFs) are created by partitioning the series. Considering the shape and amplitude features of a spectral curve, similarities can be drawn. There are various amplitude similarity measures as well as spectral similarity measures. Amplitude similarity measure methods include the calculation of various distances like Euclidean distance (ED) while spectral similarity measure methods mainly include spectral angle mapper (SAM) and spectral correlation coefficient (SCC) among others.