Kernel methods and machine learning techniques for man-made object classification in SAR images

Abstract

The image processing techniques with computer automated object recognization is an emerging area of research in several engineering and biomédical applications. The images created by Synthetic Aperture Radar (SAR) require complex image processing for intelligence extraction. A technique for man made object recognization in SAR created images is presented here. The kernel methods along with machine learning algorithms are investigated in this paper. The kernel methods allow efficient mapping from non-linear to linear feature space and integrate with several existing linear pattern matching techniques. The image's spatial characteristics are used as data for kernel functions. With MATLAB simulation results the kernel based man-made object classification is verified for different sizes of data sets under different conditions.