Paper Title
SVM Kernel Functions Analysis for Land Cover Mapping using Earth Observing-1 and Landsat 8 Data

Abstract
The Earth’s surface information provided by land cover (LC) is required in many wide-scale applications. Remotely sensed images is one of the most common applications in this branch. Different algorithms developed and applied for this purpose such as SVM. Depending on the choice of the kernel function and its parameters, classification accuracy produced by SVM variation. In this study, for LC classification of Kota Bharu district in Kelantan state, Malaysia, SVM with Radial Basis and Polynomial kernel functions and their estimated optimum parameters was applied using EO1 (ALI) and Landsat 8 data. Results showed, MLC produced higher accuracies (1%) for Landsat 8 OLI than EO-1 (ALI). SVM outperform the MLC in terms of overall and individual class accuracies especially with the use of RBF kernel produced OA 91% and 90% for the Landsat 8 OLI and the EO-1 (ALI) image respectively. SVM with RBF kernel, for all cases, exceeded SVM with polynomial kernel about 4.5 % improvement. This proves the robustness of the RBF kernel over the polynomial one. Finally, SVM are found more powerful (approximately 3.7% higher) than the MLC for the data sets. Some important findings were also obtained concerning the changes in LC in the study area Keyword - Land Cover Mapping, Support Vector Machines, Radial Basis and Polynomial Kernel Function, Maximum Likelihood