Combining one-class support vector machines and hysteresis thresholding: Application to burnt area mapping

Abstract

In this paper, we focus on burnt area mapping using a single post-fire high resolution satellite image. Concerning image classification problems, Support Vector Machines (SVM) have shown great performances. They learn how to distinguish two classes by finding the optimal hyperplane which maximizes the distance between the hyperplane and the training examples. In this paper, we propose to use the One-Class SVM algorithm, an extension of the original two-class SVM which uses only the positive examples in training and testing. This classification algorithm is then followed by a hysteresis thresholding to enhance the image segmentation. To validate the efficiency of the proposed approach, it is tested on high resolution satellite images and the results are compared to the ground truths.