Compound Gaussian Clutter(CGC) model has been widely used in fitting sea clutter with high resolution and low grazing angle. The intensity distribution of CGC with inverse Gamma texture is Generalized Pareto(GP) distribution. In real radar environment, it is very difficult to obtain large-size independent and identically distributed clutter samples due to the non-stationary and non-uniform characteristics of sea surface. Therefore, a Bayesian estimation method for parameters of GP distribution is proposed. The parameters of GP distribution with small sample are obtained by updating the prior knowledge of data online. Simulation results show that the proposed method can achieve more accurate parameter estimation compared with the conventional methods when the prior knowledge is reliable.