Classification of moving ground radar objects by using bispectrumbased features

In this paper, a novel bispectrum-based strategy applied for moving object classification by using the Doppler radar surveillance systems is proposed. Time-frequency distributions (TFDs) computed by bispectrum-based processing of signals backscattered by moving objects are used for extraction of classification features. Phase coupling contributions extracted from radar backscattering serve as the classification features contained in bispectrum-based TFDs. Coherent and homodyne Doppler radar operating at frequency of 33 GHz and continuous mode is exploited in experimental measurements. Experimental data are collected for five target classes: a single walking human, a group of walking humans, a bicyclist, a vehicle, and vegetation clutter. Classification performance of the proposed bispectrum-based automatic target recognition (ATR) radar system is examined by comparing it to the ATR systems using common energy-based TFDs. The benefits of the bispectrum-based approach are demonstrated and studied.