Dimension Reduction and Its Effects in Hyperspectral Data Classification

Abstract

Hyperspectral data imaging systems have gained significant attention from various research experts and institutions in the recent past. Hyperspectral data contains several contiguous bands which make detailed and precise classification possible. However neighbor bands are highly correlated and this makes the classification problem challenging in hyperspectral data. To overcome these difficulties one method is to reduce the dimension of the hyperspectral data. The central focus of this paper, is to analyze the effect of dimension reduction in hyperspectral data classification systems. For this purpose, a non-linear dimension reduction technique, isometric feature mapping (ISOMAP) is implemented on a well-known dataset from the literature. The Support Vector Machine (SVM) and K-Nearest Neighbors (KNN) classifiers have been utilized on both original sized dataset and reduced sized dataset to show the effectiveness of the implemented dimension reduction technique.