USE OF SVM KERNEL FOR BETTER CLASSIFICATION ACCURACY FOR REMOTE SENSING DATA

The classification of remote sensing images is a challenging task, as image contains bulk of information which is to be carefully extracted and used. Support vector machine (SVM) has proven its excellence in the field of remote sensing image classification. Using SVM as a classifier multiple kernel methods can be constructed and tested over a specified set of image dataset. In this paper, we present a framework in which multiple SVM kernels are tested on a remotely sensed image for analysis of behaviour of kernel methods over image dataset. Pre-step to image classification is feature extraction on the basis of which image is classified. In our system, the image is classified into two classes on the basis of colour feature of image. Accuracy of each kernel along with delay in image classification is calculated for the task of kernel optimization over remotely sensed image. Thus a set of best suited kernel methods can be identified over a specific dataset.