Assessment Kernel Support Vector Machines in Classification of Landuses (Case Study: Basin of Cheshmeh kileh-Chalkrod)

Classification of land use extraction always been one of the most important applications of remote sensing and why different methods are created. Over time and with greater accuracy were developed more advanced methods that increase the accuracy and the extraction classes that were closer together in terms of quality are better. SVM is one of these methods in the study of this method for the extraction of forest land, farming, pasture, and the city and its various kernel includes a linear (Linear), polynomial (Polynomial), radial (RBF) and ring (Sigmoid) were evaluated to determine the best kernel to extract these applications. The results showed the best overall accuracy and kappa coefficient, respectively polynomial of degree 5, 6 and 4 and the lowest is in the ring or Sigmoid. With increasing degree polynomial (except Grade 2) were added to the overall accuracy and kappa coefficient. Overall, we found that increasing degrees of polynomial boundary between the classes better spectral resolution and in areas that were close to be more successful. It also increases the degree polynomial caused more accurately separate the boundary between the classes. Our goal is classified when the user is using more than two degrees above polynomial (preferably 5 or 6) is recommended.