SVM Parameter Optimization using ALO for Object Based Land Cover Classification

Abstract

Machine Learning algorithms are often used to solve various kinds of data classification task. Support Vector Machine (SVM) performs better for object oriented classification of high dimensional remote sensing datasets even with minimum training samples. In order to obtain improved performance in classification, the generalization and learning ability of SVM can be enriched by proper tuning of kernel and penalizing parameters of SVM. In this methodology ALO optimizer performs the optimal searching of SVM parameter in the direction of reducing misclassification rate. The proposed approach results better SVM parameters for the significant feature sub set which characterize the Landsat image objects of the study area. Performance of ALO is compared with GA based SVM parameter optimization. Accurate thematic classification map of land cover classes of the area of study also resulted in this module.