General Linear Chirplet Transform and Radar Target Classification

In this paper, we design an attractivealgorithm aiming to classify moving targets includinghuman, animal, vehicle and drone, at groundsurveillance radar systems. The non-stationary reflectedsignal of the targets is represented with a novelmathematical framework based on behavior of thesignal components in reality. We further propose usingthe generalized linear chirp transform for the analysisstage. To enhance the classification performance, therotation invariant pseudo Zernike-Moments are extractedfrom the time-frequency map. Consequently,the obtained features are trained to the k-NN classifier.In the numerical experiments we show the superiorityof the proposed method in comparison withthe existing recent counterparts, for both performanceas well as the computational complexity. The resultsindicate that the proposed method obtains the rate of95% accuracy in classification performance, when thesignal to noise ratio is higher than 25dB. In fact,a rotating propeller on a fixed-wing aircraft, the multiplespinning rotor blades of a helicopter, or an UnmannedAerial Vehicle (UAV); the vibrations of an engine shakinga vehicle; an antenna rotating on a ship; the flappingwings of birds; the swinging arms and legs of a walkingperson; and many other sources are the source of micromotion,are known as the micro-Doppler, and can be usedfor target classification and reduction of the sensor falsealarm rate1 [7].