Suppress False Alarms by Exploiting Ambiguity Features of Targets with Machine Learning

False-alarm suppression and ambiguity resolution are two critical issues for pulse-Doppler (PD) radars. Most previous works attempted solving them independently. Besides, the short-dwell time in the real applications of search radars imposes great challenges for the false-alarm suppression methods based on time-frequency features in the previous works. This work proposes, for the first time, to leverage the ambiguity features of targets that are generated from ambiguity resolution to suppress false alarms. A machine learning model, bagged-trees, is utilized to distinguish true targets from false alarms in the feature spaces in a data-driven way. We also present a new detection paradigm of low-threshold detection followed by the proposed ML-based false-alarm suppression. Extensive filed experiments show that the new paradigm can achieve a significant improvement in the detection performance for PD radars.