An M-ary KMP classifier for multi-aspect target classification

The kernel matching pursuit (KMP) algorithm is re-formulated in the framework of the theory of optimal experiments, using a weighted sum of squared errors as the loss function, and it is extended to the case of M-ary target classification and kernel optimization. The M-ary KMP classifier is applied to the multi-aspect classification of moving targets based on high-range resolution (HRR) radar signatures, for which the target-sensor orientations are assumed approximately known. A multi-aspect processing method is presented based on the use of the estimates of target-sensor orientation angles. The KMP classification results for ten MSTAR targets are presented, with a comparison to corresponding results using the relevance vector machine (RVM).