Multi Features Advanced Support Vector Machine Method For...

Multi-feature advanced support vector machine method for classification of polarimetric synthetic aperture radar data Purnima Arora1, Dr.Paras Chawla2, Gaurav Malik3 1,2,3Electronics Communication Engineering, Seth Jai Parkash Mukand Lal Institute of Engineering Technology, Radaur, Yamunanagar, Haryana, India-135133, E-mail: 1purnima5142@jmit.ac.in; paraschawla@jmit.ac.in2; 3gauravmalikece@gmail.com Abstract— Support Vector Machine (SVM) is regarded as a good alternative of the traditional learning classification. Terrain classification using polarimetric synthetic aperture radar (POLSAR) imagery has been a very active research field over recent years. Because of support vector’s excellent learning performance, it has become a research hot spot in the field of machine learning. In this paper, we firstly evaluate and compare different features for classifying polarimetric SAR imagery and then we will describe two SVM methods: Normal represents the one without feature selection and weighted represents with optimal multi-feature selection. Obviously, the weighted SVM returns a sparse set of features and employs less SV’s than normal. However, the performance of weighted SVM is no worse and even better than the performance of normal one. The excellent recognition rates achieved in experiments indicate that advanced (weighted) SVM is well suited for SAR image cl assification. The good performance of the proposed methods is illustrated in three challenging remote sensing