Partially Supervised Classification: Based on Weighted Unlabeled Samples Support Vector Machine

Abstract

This paper addresses a new classification technique: Partially Supervised Classification (PSC), which is used to identify a specific land-cover class of interest from a remotely sensed image using unique training samples that belongs to a specified class. This paper also presents and discusses a newly proposed novel Support Vector Machine (SVM) algorithm for PSC. Accordingly, its training set includes labeled samples that belong to the class of interest and unlabeled samples of all classes randomly selected from a remotely sensed image. Moreover, all unlabeled samples are assumed to be training samples of other classes and each of them is assigned a weight factor indicating the likelihood of this assumption; hence, the algorithm is called ‘Weighted Unlabeled Sample SVM’ (WUS-SVM). Based on the WUS-SVM, a PSC method is proposed. Experimental results with both simulated and real datasets indicate that the proposed PSC method can achieve encouraging accuracy and is more robust than the 1-SVM and the Spectral Angle Mapping (SAM) method.