Model Selection of Sea Clutter Using Cross Validation Method

This work concerns a model selection of sea radar clutter used for adaptive target detection. Three distributions without thermal noise are considered; K, Pareto type II and compound Gaussian inverse Gaussian (CG-IG) with scale and shape parameters. The model selection is carried out by comparing the experimental complementary cumulative distribution function (CCDF), drawn from the recorded data intensity, to a set of the CCDF curves derived from the underling models. To do this, the cross validation technique is used after dividing a set of data into four segments. The best distribution is selected in which the mean of the means square of errors (MSEs) between the real CCDF curve and the fitted CCDF curve is minimal. Fitting comparisons of models are illustrated through overall data of Intelligent PIxel X-band radar (IPIX). From this study, it is shown that the Pareto type II distribution is suited in several cases of a low cell resolution. On the other hand, the K and CG-IG models characterize generally sea clutter with medium and high cell resolutions.