Tri-Feature-Based Detection of Floating Small Targets in Sea Clutter

It is always a challenging problem for marine surface surveillance radar to detect sea-surface floating small targets. Conventional detectors using incoherent integration and adaptive clutter suppression have low detection probabilities for such targets with weak returns and unobservable Doppler shifts. In this paper, three features of a received vector at a resolution cell-the relative amplitude, relative Doppler peak height, and relative entropy of the Doppler amplitude spectrum-are exploited to give returns with targets from sea clutter. Real datasets show that each feature alone has some discriminability, and the three features jointly exhibit strong discriminability. Due to diversity of targets in practice, it is impossible to get features of returns with all kinds of targets. We recast detection of sea-surface floating small targets as a one-class anomaly detection problem in the 3D feature space. A fast convexhull learning algorithm is proposed to learn the decision region of the clutter pattern from feature vectors of clutter-only observations. As a result, a tri-feature-based detector is developed. The experiment results for the IPIX datasets show that the proposed detector at an observation time of several seconds attains better detection performance than several existing detectors.