Machine Learning based Non-Cooperative Target Recognition with Dynamic RCS Data

There is ever increasing requirement of identification of non-cooperative aerial targets for ensuring flight safety, security, and effective airspace management. The identification or classification can be done by utilizing target features and estimated Radar Cross Section (RCS) obtained from modern multifunction phased array radars. The estimated RCS is dependent on various governing factors, and it is significantly stochastic in nature. In this paper authors propose to implement machine learning algorithms on the dynamic RCS for radar target identification. We have considered three fixed wing aerial targets of sizes 3.7 m, 7.5 m and 14.9 m for classification. Time varying RCS or dynamic RCS for these targets has been generated based on aspect angle derived from flight path as reported by Automatic Dependent Surveillance-Broadcast (ADS-B) data source and interlinking the same with EM solver generated RCS value. The classification accuracy of 97.4 %, 63.5% and 66.3 % has been obtained while using Random Forest, Support Vector Machine and Adaptive Boosting Algorithm respectively.