UNSUPERVISED ONE-CLASS SVM USING A WATERSHED ALGORITHM AND HYSTERESIS THRESHOLDING TO DETECT BURNT AREAS
Abstract
This paper addresses the issue of color image classification. Support Vector Machines (SVM) have shown great performances concerning classification problems but require positive and negative training sets. One-Class SVM allow to avoid the negative training set choice. We also propose to automatically select the positive training set by using the watershed algorithm on the 3-D histogram. Finally a hysteresis thresholding allow to improve the cluster edges. Our method is applied to multispectral satellite images in order to assess burnt areas after a forest fire. The results are compared to official ground truths to validate the approach.
Origin : Files produced by the author(s)
Loading...