Characteristic function based method for SVM classification of maneuvering over the horizon targets

In this paper, we propose a new classification method based on characteristic function (CF) and support vector machine (SVM). In order to validate the new approach, we classify three groups of airborne over-the-horizon radar (OTHR) targets. Since signal models make the basis for analysis and enhancement of OTHR performance, choosing an appropriate model has always been a matter of concern. On the other hand, the returned signal from a maneuvering target is more often a multi-component signal with time-varying frequency, hence, we model the received signal as being comprised of a chirp faded by the radar cross section (RCS) plus Gaussian white noise and K-distributed (un)correlated clutter. Little work has been done on OTHR target classification. In order to assess the new classification approach based on CF, we compare our method with discriminant analysis (DA), decision tree (DT), and multi-layer Perceptron neural network (NN). It will be depicted through extensive simulations that the proposed CF and multi-phase SVM method's error in classifying airborne targets is about 3.5% less than existing classification methods'.