A classification system for remote sensing satellite images using support vector machine with non-linear kernel functions

The growing productions of maps are generating huge volumes of data that exceed people´s capacity to analyze them moreover these data sets have different resources and types. It seems appropriate to apply knowledge discovery methods like data mining to spatial data so, one of the most significant application in spatial data mining is classification for remote sensing images. This paper proposes a classification system for remote sensing ASTER satellite imagery using SVM with non-linear kernel functions. The proposed system starts with the identification of selected area of study. This is followed by a preprocessing phase to enhance the quality of the input remote sensing satellite image and to reduce speckle without destroying the important features using mapping polynomial algorithm as geometric correction. Followed by, applying threshold algorithm for image segmentation. Then features are extracted using object based algorithm. Followed by, image classification using SVM with nonlinear kernel function. It is tested and evaluated on selected area of interest in the north-eastern part of the Eastern Desert of Egypt (Halaib Triangle). The obtained results carried out that SVM with RBF kernel function has the highest classification accuracy ratio.