Improved SAR Vessel Detection Based on Discrete Texture

Abstract:
The conventional model to encompass heterogeneity in sea clutter in SAR images is the product or compound model, which modulates the Gaussian-distributed homogeneous clutter with a statistically independent texture random variable. If the texture is assumed to follow a γ-distribution, the compound model becomes K-distributed. The K-distribution, however, omits the receiver noise and is not felt to be suciently robust to cover the range of environments that are expected practically. This paper proposes a new sea clutter model based on the idea that the sea texture can be statistically modelled as discrete in nature instead of the status-quo, i.e. continuous texture statistics. The proposed model also correctly accounts for the additive thermal noise. The algorithms to compute the required texture parameters and subsequently the detection threshold stand out through their numerical simplicity, enhancing robustness and reducing the computational complexity considerably. The theoretical ndings are corroborated by experimental RADARSAT-2 data.