Spectral noise reduction and smoothing using local cubic least square regression from hyperion reflectance data

Hyperion data contain significant amount of spectral noise even after radiometric and spectral calibration, noise reduction and atmospheric corrections. The noise appears in the form of false absorption features which can potentially mislead spectral analysis. In this paper, we present a hybrid approach for removing spectral noise from Hyperion hyperspectral reflectance imagery. In this study, an MNF transformation and low-pass filtering are used in tandem to reduce random noise, and a local cubic least square regression based algorithm (LCLSR) is used in spectral domain to estimate a common spectral gain factor for each pixel's spectra in an image to get smooth reflectance spectra.