Removal of Cosmic Spikes from Hyper-Spectral Images Using a Hybrid Upper-Bound Spectrum Method

Abstract

A hybrid of the upper-bound spectrum (UBS), principle component analysis (PCA), and median filter (MF) algorithms is used to efficiently remove cosmic spike spectral artifacts from hyper-spectral data matrices (DM). The resulting UBS-DM method is shown to introduce less spectral distortion than the widely used MF cosmic spike suppression algorithm alone. The new method relies on a truncated series of PCA eigenvectors to produce an upper-bound spectrum, which is in turn used to detect and remove cosmic spikes. The PCA eigenvectors are separated into three groups, each of which is treated in a different manner. The first group of eigenvectors, which contain most of the spectral correlation of interest, are not preprocessed prior to applying the UBS-DM method. The second group, which are more severely contaminated by cosmic spikes (but may also contain important spectral correlations), are MF preprocessed before applying UBS. The third group of eigenvectors, which contain virtually no spectral information, are ignored in regenerating the upper-bound spectra used for spike detection and removal. The UBS-DM algorithm is tested using both synthetic and experimental (iron oxide) Raman hyper-spectral data sets, each containing several thousand spectra, with about 1000 wavelength pixels.