A Comparative Study of Land Cover Mapping Based on Support Vector Machine

Abstract

The richness of remote sensing images in information, due to the number of bands, makes them widely used in detecting and classifying terrestrial objects. The purpose of this study is a classification of multispectral images for mapping land occupation in the Mohammadia region (located in the west of Algeria). We developed a comparative study on the classification of the study subject image using the three following kernel functions: linear (LN), polynomial (PL), and radial basis function (RBF). After selecting the desired bands of the multispectral image, the training of the SVM is then carried out on the seven interest zones of the studied region: buildings, dense vegetation, sparse vegetation, forest, bare land, and roads. The obtained results are very promising, where the best classification rates were obtained by the use of the RBF kernel (97.91%) and the polynomial kernel (98.79%).