Hyperspectral data discrimination based on Ensemble Empirical Mode Decomposition

The classification of hyperspectral data is an important issue. This investigation adopts a novel hyperspectral data classification approach using Ensemble Empirical Mode Decomposition (EEMD). First, the EEMD is applied to decompose the spectra into several components. Then, some selected components are applied to generate the classification indices. The classification indices include correlation coefficients, weighted Euclidean distance and weighted absolute distance. Two spectrum data sets are selected in the experiment. The first concerns vegetation while the other is about soils. The experiment results demonstrate that EEMD can characterize the spectral properties. Moreover, the decomposed components are able to separate the spectrum data when different indices are applied. The proposed method enhances hyperspecral data discrimination of different classes. The recognition rate are from 8.00% to 195.33%, 37.53% to 531.37%, and 26.31% to 423.84%; and are measured by correlation coefficients, weighted Euclidean distance and weighted absolute distance, respectively.