On computing first and second order derivative spectra

Enhancing resolution in spectral response and an ability to differentiate spectral mixing in delineating the endmembers from the spectral response are central to the spectral data analysis. First and higher order derivatives analysis of absorbance and reflectance spectral data is commonly used techniques in differentiating the spectral mixing. But high sensitivity of derivative to the noise in data is a major problem in the robust estimation of derivative of spectral data. An algorithm of robust estimation of first and second order derivative spectra from evenly spaced noisy normal spectral data is proposed. The algorithm is formalized in the framework of an inverse problem, where based on the fundamental theorem of calculus a matrix equation is formed using a Volterra type integral equation of first kind. A regularization technique, where the balancing principle is used in selecting a posteriori optimal regularization parameter is designed to solve the inverse problem for robust estimation of first order derivative spectra. The higher order derivative spectra are obtained while using the algorithm in sequel. The algorithm is tested successfully with synthetically generated spectral data contaminated with additive white Gaussian noise, and also with real absorbance and reflectance spectral data for fresh and sea water respectively.