Wavelet Analysis of Hyperspectral Reflectance Data for Spectral Feature Extraction

摘要

This study determined the potential of wavelet-based analysis for extracting spectral features of hyperspectral reflectance signals. The dyadic discrete wavelet transform is proposed for feature extraction from a high dimensional data space. The wavelet's inherent multi-resolution properties are discussed in terms related to multi-spectral and hyperspectral remote sensing. Wavelet can focus on the local structure of the signal through adjusting the scale parameter in the course of focusing. So we can find the singularities and the inflexions of the original signal. The absorption strips are thus detected consequently with the local wavelet transform modulus (absolute value) maxima. The results show a superior performance of the proposed wavelet-based features that are more meaningful for spectral feature extraction when compared to conventional methods.