Maritime Radar Target Detection Using Convolutional Neural Networks

Maritime airborne surveillance radars operating at medium to high grazing angles experience significant sea clutter returns with a time and range-varying Doppler spectra, thus making the detection of small targets extremely difficult. This has led to the development of new techniques to better isolate targets from the background sea clutter. In this paper, two convolutional neural network approaches are proposed. They include a method of reconstructing maritime point target signals, while rejecting non-homogeneous clutter and a novel approach that is able to directly locate the target signal peak from sea clutter using a convolutional autoencoder trained in a supervised manner. These two approaches may be considered pre-processing stages in a detection scheme, with a simple threshold detection scheme used to detect targets with a desired false alarm rate.