Multisensor fusion for seabed classification

Automatic seabed classification can be achieved using acoustic sensors but methods need to be improved. In order to get better classification reliability, we propose to use complementarity between several acoustic sensors: normal incidence echo sounder, sidescan sonar. The new feature is that the sonar (Klein), provides a high resolution sidescan sonar image which pixels are colocated with high resolution bathymetric points. After extracting information from each of these sources, the key point is to fuse them to be able to classify the seabed. We propose to compare several fusion approaches: signal-level fusion based on multidimensional classification algorithms, and a symbol-level fusion based on the Dempster-Shafer evidence theory. These methods are tested on real sonar data.