Hyperspectral Image Classification using Digital Signature Comparison based Classifier

Abstract

In this paper, a pixel-based supervised classifier based on Digital Signature Comparison (DSC) for hyperspectral image classification is proposed. The classifier is conceptually simple, easy to implement, and requires less memory space for storing training parameters when compared to the widely used and popular Support Vector Machine (SVM) classifier and the Naive Bayes (NB) classifier. In the proposed classifier, the digital signatures are generated by successive comparison of the Digital Number (DN) of the present spectral band to the DN of its adjacent spectral band from the set of original spectral bands of the hyperspectral image, for both training data (for each class label) and test data. Then, a comparison of the digital signatures of training data with test data and finding the number of matches of comparison are performed to assign a test pixel to the class for which it has the highest or majority of votes (or matches). Performance accuracy derived through simulation results on two hyperspectral images, namely Washington-DC Mall (WDC-M) and Salinas-A, substantiate the effectiveness of the proposed classifier (with lesser training parameters’ storage).