Feature Fusion for Classification Enhancement of Ground Vehicle SAR Images

Abstract

In this paper four feature extraction techniques are utilized to extract features from Synthetic Aperture Radar images namely as Radial Harmonic Fourier Moment, Local Binary Pattern, Haar Wavelet and Radon Transform. Holdout, 2-fold and 10-fold cross validation techniques are used for classification of images by using Support Vector Machine classifier. Haar Wavelet and Radon Transform does not reduce the dimensions of input data, hence Principle Component Analysis is applied to reduce the dimensionality. Fusion is established by concatenation of all the features of Radial Harmonic Fourier Moment, and Local Binary Pattern and selected features of Haar Wavelet Radon Transform. Experimental results verify that fused technique represents an improvement in accuracy.