Radar Target Recognition Based on Modified Characteristic Polarization States

This paper proposes a radar target recognition algorithm based on a feature set extracted from the target characteristic polarization states (CPS) and evaluated at a set of target resonant frequencies in the frequency domain. The algorithm involves measuring the proximity between training (stored) prototypes and a test prototype. For this task, the algorithm implements the nearest neighbour (NN) algorithm at each resonance to give a vote, and then the majority resonance vote is used to identify the test target prototype. The approach begins by using the power optimization procedure of the Kennaugh power matrix at a set of resonant frequencies to derive a series of CPS (expressed in normalized Stokes vector form) together with their associated receiver power levels. Then, certain CPS are selected and weighted by their associated power levels to create the feature set. Finally, a Euclidean metric is used to determine the minimum k-NN distances between the training prototypes and a test prototype, and the target selected by the majority of the resonances is the identified target. The paper includes the identification performance for two symmetrical targets of similar electrical lengths.