Machine Learning Techniques for Coherent CFAR Detection Based on Statistical Modeling of UHF Passive Ground Clutter

Abstract

Ultra high frequency (UHF) passive ground clutter statistical models were determined from real data acquired by a passive radar for the design of approximations to the Neyman–Pearson detector based on machine learning techniques. The cross-ambiguity function was the input space without any preprocessing. The Gaussian model was proved to be suitable for high Doppler values. Other models were proposed for Doppler close to zero, where ground clutter and low bistatic Doppler targets concentrate. Likelihood ratio detectors were built for this Doppler region, and a neural-network-based adaptive threshold technique was designed for fulfilling false alarm requirements throughout all the input space. The proposed scheme outperformed a conventional passive radar one and could be used as a reference for future designs.