Robust Detection in Compound Gaussian Clutter Based on Bayesian Framework

The texture component of compound Gaussian model determines the non-Gaussian characteristics of clutter, and the uncertainty of the texture component can result to the detection performance degradation of the conventional detectors. In this paper, based on the Bayesian framework, the prior distribution is used to denote the uncertainty of texture component, and the impact of the prior model on the robust detection performance is discussed. Two kinds of prior models are considered: non-informative prior model and the informative prior model. Non-informative prior models include the Jeffery prior model and generalized non-informative prior model, and the Normalized Matched Filter (NMF) is given using these prior models. Conjugate prior distribution is used as informative prior model, and Knowledge Aided NMF (KA-NMF) is given. The structure and threshold of KA-NMF are the function of the parameters of prior model. In this paper, the sensitivity of the detection performance of KA-NMF to the parameters of prior model is analyzed. Further more, the non-informative prior model is used to denote the parameters, and the Hierarchical Bayesian NMF (HB-NMF) is given. The computer simulation and real sea clutter data analysis results show that, the HB-NMF detection performance has no relation with the parameters of prior model, and its robustness and detection performance outperform the KA-NMF and NMF respectively.