Hybrid Features-Based Efficient Radar Target Classification Using Support Vector Machine

Abstract

Automatic classification of beyond-visual-range (BVR) areal targets plays an important role in early awareness and alertness of air threat. Various target classification approaches, namely target kinematics, radar cross-section (RCS) of target, Doppler and high-resolution range profile (HRRP)-based techniques, are available in the literature. However, these approaches consider one type of target signature, i.e. either target kinematics or Doppler or RCS or HRRP, and hence, the classification accuracy achieved with these techniques is limited. The classification accuracy may get improved by considering combination of target signatures together. In this article, we propose hybrid features-based efficient radar target classification using support vector machine (SVM), wherein the combination of the target signatures is considered together. The multi-aspect target signatures include kinematic parameters (KP) of the target which are obtained using robust interactive multiple model (IMM)-based tracking algorithm and aspect-specific target strength (ASTS) of the target which is obtained from signal processor (SP). A multi-class (3-class) SVM classifier is envisaged for classifying multiple aerial targets, namely commercial, fighter and trainer jet. Initially, the SVM classifier is trained using hybrid features extracted from the trajectories of three aircrafts said above. Subsequently, it was used to predict the class label for the test samples. We compared the performance of classifier with hybrid features versus KP and found that hybrid features produced better results.