General Linear Chirplet Transform and Radar Target Classification

In this paper, we design an attractive algorithm aiming to classify moving targets including human, animal, vehicle and drone, at ground surveillance radar systems. The non-stationary reflected signal of the targets is represented with a novel mathematical framework based on behavior of the signal components in reality. We further propose using the generalized linear chirp transform for the analysis stage. To enhance the classification performance, the rotation invariant pseudo Zernike-Moments are extracted from the time-frequency map. Consequently, the obtained features are trained to the k-NN classifier. In the numerical experiments we show the superiority of the proposed method in comparison with the existing recent counterparts, for both performance as well as the computational complexity. The results indicate that the proposed method obtains the rate of 95% accuracy in classification performance, when the signal to noise ratio is higher than 25dB. In fact, a rotating propeller on a fixed-wing aircraft, the multiple spinning rotor blades of a helicopter, or an Unmanned Aerial Vehicle (UAV); the vibrations of an engine shaking a vehicle; an antenna rotating on a ship; the flapping wings of birds; the swinging arms and legs of a walking person; and many other sources are the source of micromotion, are known as the micro-Doppler, and can be used for target classification and reduction of the sensor false alarm rate.