Decision Fusion for Hyperspectral Image Classification Based on Minimum-Distance Classifiers in the Wavelet Domain

W. Li, S. Prasad, E. W. Tramel, J. E. Fowler, & Q. Du

• Link
• PDF

Astract

A decision-fusion approach is introduced for hyperspectral data classification based on minimum-distance classifiers in the wavelet domain. In the proposed approach, multi-scale features of each hyperspectral pixel are extracted by implementing a redundant discrete wavelet transformation on the spectral signature. Following this, a pair of minimum distance classifiers-a local mean-based nonparametric classifirer and a nearest regularization subspace-are applied on wavelet coefficients at each scale. Classification results are finally merged in a multi-classifier decision-fusion system. Experimental results using real hyperspectral data demonstrate the benefits of the proposed approach-in addition to improved classification performance compared to a traditional single classifier, the resulting classifier framework is effective even for low signal-to-noise-ratio images.

BibTeX

    inproceedings{lpt2014,
    Address = {Xi'an, China},
    Author = {Wei Li and Saurabh Prasad and Eric W. Tramel and James E. Fowler and Qian Du},
    Booktitle = {IEEE China Summit on Signal and Info. Processing},
    Month = {July},
    Pages = {162--15},
    Title = {Decision Fusion for Hyperspectral Image Classification Based on Minimum-Distance Classifiers in the Wavelet Domain},
    Year = {2014}}