Improved covariance matrix estimation in spectrally inhomogeneous sea clutter with application to adaptive small boat detection

Asymptotically optimal coherent detection techniques yield sub-clutter visibility in heavy-tailed sea clutter. The adaptive linear quadratic detector inherently assumes spectral homogeneity for the reference window of the covariance matrix estimator. This paper investigates the validity of this assumption on real data that have been recorded with a medium resolution X-band radar. Proving empirically that this basic assumption is the exception rather than the rule, the effects of spectral inhomogeneity are investigated. Various improvements to the current estimator and detector are suggested and evaluated. This is specifically applied to the detection of small rigid inflatable boats, with empirical results presented for a range of boat manoeuvres. A specific example highlights the vulnerability of the chosen detector to self-masking.