MULTISCALE FEATURE EXTRACTION OF HYPERSPECTRAL DATA USING WAVELET TRANSFORM

摘要

The purpose of feature extraction is to abstract substantial information from the original data inputrnand filter out redundant information. In this study, we transfer hyperspectral data from thernoriginal-feature space into a scale-space plane using the wavelet transform to extract the significantrnspectral features. The wavelet transform can focus on localized signal structures with a zoomingrnprocedure. The absorption bands are thus detected with the wavelet transform modulus maxima at eachrnsingularity point of the spectral curve from the decay of the wavelet transform amplitude. The localrnfrequency variances provide some useful information about the oscillations in the hyperspectral curvernfor each pixel. Various types of materials can be distinguished by the differences in the local frequencyrnvariation. This new method generates more features that are meaningful and is more stable than otherrnknown methods for spectral feature extraction.