• adaptive detection
• compound-Gaussian model
• HFSW radar
• sea clutter
• statistical analysis

This thesis concerns with the analysis of data collected by two low-power surface wave
(SW) over-the-horizon (OTH) radars. The two systems, developed at the University of Hamburg
and named Wellen Radar (WERA), concurrently operate in a real maritime surveillance
scenario in the Bay of Brest (France), with the double purpose of both sea current sensing
and vessel detection. The research activity presented in this dissertation develops following
the second topic and is interested in the statistical analysis of recorded data and then in the
consequent ship detection.
Part I deals with the statistical analysis of sea clutter in the HF band. Detailed investigations
have been carried out on which is the best statistical model. Numerical results
have shown interesting features, both before and after beamforming. The clutter amplitude
is Rayleigh or Weibull distributed for the majority of near and middle-range cells, while it
cannot be fitted very accurately by the most common amplitude distributions at long distances.
This is caused by the highly impulsive nature of interferences, both man-made and
natural, which mask the clutter signal. The resulting signal, however, can still be modeled
as a compound-Gaussian (CG) process, as verified by an in depth analysis of the probability
density functions (PDFs) of both the speckle and texture signal components. The spectral
features of the signal have been studied as well. They demonstrated to be a precious tool for
recovering information about the nature of non-Gaussianities. Finally, the signal spectrum has
been described as an auto-regressive (AR) model.
Part II is instead focused on the application of decision-based techniques. Starting from
the previous results, our investigation develops on the analysis of detection algorithms based
on the normalized adaptive matched filter (NAMF) and the impact of clutter non-stationarities
on detection performance. To deal with the unknown clutter statistics, three covariance matrix
estimators, namely the sample covariance matrix (SCM), the normalized sample covariance
matrix (NSCM) and the fixed point covariance matrix (FPCM), or approximate maximum
likelihood (AML), are investigated and their performances evaluated. The idea is to capitalize
these results and exploit WERA for detecting ships well beyond the horizon in a complex
integrated maritime surveillance (IMS) scenario. The final superposition of WERA detection
maps and automatic identification system (AIS) ground truth data will allow to quantify the
goodness of the algorithms. In this sense, guidelines about future work are provided as well.