Investigating the Performance of Metaheuristic Population-Based Algorithms to Optimize the Parameters of Support Vector Machines in Classification of Polarimetric Images

Because of using the different polarization of electromagnetic wave in Polarimetric imagery، it provides a rich source of information from the several aspects of targets. Recently، Polarimetric images as a powerful and efficient tool have been interested to identify the various objects in the complex geographic areas. In order to extracting information، classification of Polarimetric image has an important effect. Support Vector Machines (SVMs) due to their operation based on geometrical characteristics and robustness in high dimensional space، are considered as a suitable case for classification of Polarimetric images. However، the performance of SVMs classifier is strongly influenced by its parameters. Therefore، the optimum values for SVMs parameters should be determined to achieve SVMs classifier with maximum efficiency. Traditional optimization techniques because of computational complexities in the large search space usually trap in local optimum. Thereby، it is inevitable to apply Meta -heuristic Algorithms which performe exploration and exploitation to obtain global optimum. In this paper، the potential of Genetic، Bees and Particle Swarm Optimization (PSO) Algorithms as powerfull techniques in determining the optimum SVMs parameters are evaluated. Comparing the results، demonstrates the superior performance of PSO Algorithm in terms of classification accuracy and speed of convergence.