Invariant Parameter Estimation across Varying Seabeds in Synthetic Aperture Sonar Imagery

Side-look synthetic aperture sonar (SAS) can produce very high quality images of the seabed. The imagery generated by SAS sensors display the seafloor and associated textures, such as sand ripples or seagrass. The characterization of SAS imagery is important for applications such as the detection and classification of targets, seabed segmentation and classification, and others. In this paper, we seek to estimate invariant features of the seafloor to describe and distinguish between seafloor types. Our approach makes the assumption that the local characteristics of the true bathymetry of the seafloor can be represented by a Gaussian Markov Random Field (GMRF). Since we characterize the bathymetry of the seafloor, this method yields GMRF parameters that are invariant to acoustic sensing modality and geometry.