Deep Learning for Radar Target Detection in Non-Homogeneous Clutter

Constant false alarm rate (CFAR) detectors are widely used in radar systems for detecting target returns against a background with thermal noise, clutter and interference. Many different adaptive CFAR detector schemes are already in use, however none prove to be optimal considering the presence of non-homogeneous background environments. This paper proposes an alternative detector scheme through deep learning, showing that a deep unfolded model-based network architecture significantly outperforms conventional cell averaging (CA) CFAR, as well as standard deep convolutional networks, under challenging clutter and interference conditions.