Wavelet-based de-noising for derivative spectra analysis

Derivative analysis is one of the techniques that is suitable for the analysis of high spectral resolution data such as that derived from airborne hyperspectral sensors and field spectrometers. The use of derivative analysis provides several advantages that facilitate the extraction of information from the data. However, the derivatives of a reflectance spectrum are significantly noisier than the original spectral reflectance curve. The advantages of derivatives are therefore offset by the introduction of such noise. A number of methods for de-noising signals have been used in the past. Our method is based on the use of wavelets. In this paper, a technique of de-noising spectra using the discrete wavelet transform is described. The de-noised derivative spectra are then used in a template-matching scheme, with image endmembers providing the templates. The result is an initial 'hard' classification of part of the study area in Central Spain using DAIS 7915 airborne hyperspectral data.