Multisensor integration for underwater scene classification

A new approach is developed for automated classification of seafloor scenes using sonar sensors and optical imagery. The sonar image yields backscatter estimates, while the optical stereo imagery yields surface roughness parameters. This information is combined and used in a composite roughness model of high-frequency bottom backscattering phenomenon. The model provides estimates of material density ratio and sound velocity ratio for the seafloor. These parameters serve as physically meaningful features and enable classification of the seafloor. Experiments on real data illustrate the classification of imaged seafloor as sandy sediment or sediment rich in manganese nodules.