Automatic recognition of seafloor features in sub-bottom profiles using eigenimages

Senior thesis written for Oceanography 445

[author abstract] Sonar is the primary tool used to remotely survey the seafloor. Sub-bottom profilers use pulses of sound to penetrate the seafloor and create an acoustic profile of the seafloor and sub- bottom. The profiles are traditionally inspected by sight to identify features of interest which is time-consuming and requires an experienced human’s eye. I have developed a method using eigenimage analysis that can automatically distinguish between three different types of seafloor features found in the fjords of Nootka Sound: Sediment, sills, and rockslides. The method uses a training set of features to build eigenimages that represent the orthogonal variance between different features. Test features are projected onto the eigenimages and compared to the average three feature types. The resulting projection coefficients are a signature characteristic to each feature type and can be used to identify and classify seafloor features.

University of Washington School of Oceanography