A Novel Curve Fitting Feature Extraction Method for Hyperspectral Image

In hyperspectral image classification, many of the existing feature extraction methods using spectral information have aroused extensively attention. However, it is difficult to characterize the geometric properties of the spectral response curves (SRCs) only depending on the spectral information. A novel feature extraction method using Maclaurin series function curve fitting was proposed in this paper. The new features for each spectral response curve of hyperspectral image pixels can be reconstructed through curve fitting. Then, the coefficients of the fitted Maclaurin series function are considered as extracted features that can better capture the intrinsic geometrical nature of spectral response curves. The proposed method concentrates on the reflectance coefficients information commendably that has not been addressed by lots of other analysis methods. The proposed method shows better superiority compared to conventional feature extraction methods when a maximum likelihood classifier (MLC) is used in hyperspectral image dataset Indian Pines.