11.07.2015 Views

Principles of Modern Radar - Volume 2 1891121537

Principles of Modern Radar - Volume 2 1891121537

Principles of Modern Radar - Volume 2 1891121537

SHOW MORE
SHOW LESS
  • No tags were found...

Create successful ePaper yourself

Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.

14.4 Synthetic Aperture <strong>Radar</strong> 651they assume the observation parameter distributions approximate a mixture <strong>of</strong> Gaussianor autoregressive densities closely [59].14.4.4.16 Support Vector MachineSupport vector machines (SVMs) have been shown to have strong classification, faulttolerance, and generalization abilities [61]. SVM is a margin-based classifier that uses asubset <strong>of</strong> training samples to find support vectors giving the maximum decision boundaryhyperplane <strong>of</strong> the classifier [62, 41]. For an SVM to discriminate between more than twoclasses, a parallel, cascading, or hierarchical decision scheme is generally employed [63].SVMs cannot select an optimal feature subset from the complete extracted feature set; anoptimal feature subset simplifies classification and can improve classification performance[61]. By combining an SVM with a rough set, or a rough estimate <strong>of</strong> the feature setconsisting <strong>of</strong> lower and upper approximations, the classification phase <strong>of</strong> ATR can beperformed well [61].14.4.4.17 Correlation FilterCorrelation filters are used in the classification stage <strong>of</strong> SAR ATR because they are shiftinvariant, <strong>of</strong>fering varying degrees <strong>of</strong> distortion tolerance, and are not tailored to a particularsensor or image properties [64]. Other advantages <strong>of</strong> advanced correlation techniquesinclude: filter synthesis can be performed <strong>of</strong>f-line; image preprocessing is unnecessarybecause <strong>of</strong> the filter’s distortion tolerance and robustness; the processing complexity dependson the scene size but not its content; design optimization and performance boundingis relatively easy; and the algorithm maps well into hardware [64]. We discuss only a few<strong>of</strong> the many examples <strong>of</strong> these filters below.14.4.4.18 Maximum Average Correlation HeightThe maximum average correlation height (MACH) filter is an attractive correlation filteroption because <strong>of</strong> its better distortion tolerance than other alternatives; however, it haspoor clutter rejection performance because it relies strongly on the average training image.Variations <strong>of</strong> the extended MACH filter can be used to improve the clutter rejection performance[65]. The MACH filter generates peaks from which a peak-to-sidelobe ratio (PSR) iscalculated; the test image belongs to the class with the highest PSR [64]. A MACH filter canbe used in conjunction with a distance classifier correlation filter (DCCF) to improve performanceby applying the DCCF if the MACH PSR is above an acceptable threshold [66].14.4.4.19 Distance Classifier Correlation FilterThe distance classifier correlation filter (DCCF) classifies targets using the distance <strong>of</strong>the test image correlation to the class average correlation, instead <strong>of</strong> only the outputcorrelation peak, and has shown better classification results than other correlation filters[66]. For best performance, interclass distance should be maximized and the averagecorrelation energy minimized, while the average intraclass distance, represented by theaverage similarity measure, should be minimized [66]. The DCCF involves processingby a transformation filter to maximize the average spectral separation between classesand, thus, enhance discrimination; then, the shift-invariant minimum mean square erroris calculated to determine the distance between the transformed test image and eachtransformed reference image [66]. The image is classified according to the class <strong>of</strong> thereference image yielding the smallest distance [66].

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!