Spectral library pruning based on classification techniques

Spectral unmixing is an active research area inremote sensing. The direct use of the spectral libraries in spectralunmixing is increased by increasing the availability of thelibraries. In this way, the spectral unmixing problem is convertedinto a sparse regression problem that is time-consuming. This isdue to the existence of irrelevant spectra in the library. So thesespectra should be removed in some way. In this paper, a machinelearning approach for spectral library pruning is introduced. Atfirst, the spectral library is clustered based on a simple andefficient new feature space. Then the training data needed tolearn a classifier are extracted by adding different noise levels tothe clustered spectra. The label of the training data is determinedbased on the results of spectral library clustering. After learningthe classifier, each pixel of the image is classified using it. Forpruning the library, the spectra with the labels that none of theimage pixels belong to, are removed. We use three classifiers,decision tree, neural networks and k-nearest neighbor todetermine the effect of applying different classifiers. The resultscompared here show that the proposed method works well innoisy images.

کلیدواژه ها:

hyperspectral image ، spectral library ، sparse unmixing ، machine learning