A Novel Ranking-Based Clustering Approach for Hyperspectral Band Selection

PROJECT TITLE :

A Novel Ranking-Based Clustering Approach for Hyperspectral Band Selection

ABSTRACT:

Through imaging the same spatial area by hyperspectral sensors at completely different spectral wavelengths simultaneously, the acquired hyperspectral imagery often contains hundreds of band pictures, which give the possibility to accurately analyze and identify a ground object. But, because of the problem of getting sufficient labeled coaching samples in observe, the high number of spectral bands unavoidably results in the problem of a “dimensionality disaster” (conjointly known as the Hughes phenomenon), and dimensionality reduction ought to be applied. Regarding band (or feature) selection, conventional strategies opt for the representative bands by ranking the bands with defined metrics (such as non-Gaussianity) or by formulating the band selection problem as a clustering procedure. Because of the various but complementary benefits of the two kinds of ways, it will be beneficial to use each methods along to accomplish the band selection task. Recently, a quick density-peak-based mostly clustering (FDPC) algorithm has been proposed. Based mostly on the computation of the local density and therefore the intracluster distance of every point, the product of the two factors is sorted in decreasing order, and cluster centers are recognized as points with anomalously giant values; hence, the FDPC algorithm will be thought-about a ranking-based mostly clustering methodology. In this paper, the FDPC algorithm has been enhanced to form it appropriate for hyperspectral band selection. 1st, the ranking score of each band is computed by weighting the normalized native density and the intracluster distance rather than equally taking them under consideration. Second, an exponential-primarily based learning rule is used to adjust the cutoff threshold for a totally different variety of selected bands, where it's mounted within the FDPC. The proposed approach is thus named the enhanced FDPC (E-FDPC). Furthermore, an efficient strategy, that is named the isolated-purpose-stopping criterion, is developed to automatically verify the acceptable- variety of bands to be selected. That is, the clustering method can be stopped by the emergence of an isolated purpose (the only purpose in one cluster). Experimental results on three real hyperspectral knowledge demonstrate that the bands selected by our E-FDPC approach could achieve higher classification accuracy than the FDPC and other state-of-the-art band choice techniques, whereas the isolated-purpose-stopping criterion is a reasonable method to see the preferable number of bands to be selected.

Did you like this research project?

To get this research project Guidelines, Training and Code... Click Here