Spectral feature extraction based on Orthogonal Polynomial Function fitting

We propose a new method for spectral feature extraction based on Orthogonal Polynomial Function (OPF) fitting. Given a spectral signature, it is firstly divided into spectral segments by a splitting strategy. All segments are fitted by using OPF respectively. The features of input spectrum are selected from the fitting coefficients of all segments. 10 laboratory spectra of various materials are selected to validate the ability of the proposed method. The results show that our method can efficiently mine geometric structural information of spectral signatures, and compress them into a few parameters. These parameters can be used to sparsely represent the input spectra and also well discriminate different spectral signatures. The proposed method is more powerful than the inverse Gaussian function model as it can not only will fit the red-edge spectral segment but also can fit other types of spectral curves. Also, the extracted features are slightly better than the original bands at the ability of discrimination in terms of RSDPW in Euclidean space while largely reduce the number of features. Overall, the proposed method has promising prospects in hyperspectral data analysis.