Artificial intelligence techniques for small boats detection in radar clutter. Real data validation

Artificial intelligence techniques were applied for detecting small moving targets in maritime clutter environments. Neural detectors are considered to approximate the Neyman-Pearson (NP) in composite hypothesis testing problems. Sub-optimum approaches based on the Constrained Generalized Likelihood Ratio (CGLR) were analysed, and compared to conventional implementations based on Doppler filtering that are designed to filter clutter and improve the Signal-to-Interference Ratio, and Constant False Alarm Rate techniques. The CGLR performance was significantly better at the expense of a high computational cost. As a solution, neural network training sets were designed for approximating the NP detector. The detection of small boats in Gaussian clutter was the defined case study in order to assume the design hypothesis of the conventional solutions and to study their performance under their most favourable conditions. Detection schemes were evaluated using real radar data. Neural solutions based on Second Order Neural Networks provide the best results, being able to approximate the CGLR with a significantly low computational cost compatible with real-time operations.