Bottom classification: operational results from QTC VIEW

• Main Authors: Prager, B.T., Caughey, D.A., Poeckert, R.H.
• Format: Conference Proceeding
• Language: English
• Subjects: Feature extraction; Hardware; Image sampling; Marine; Marine animals; Principal component analysis; Remotely operated vehicles; Shape; Signal processing; Signal processing algorithms; Underwater acoustics
• DOI: 10.1109/OCEANS.1995.528859

A practical seabed classification system which can operate with most echo sounders has been developed. The system consists of digital signal processor-based hardware executing statistical, energy, spectral and wavelet algorithms used to extract a feature vector whose elements are related to the shape of the seabed echo. Principal component analysis is used to create a set of weights which are used to linearly combine the elements of the full feature vector to create a reduced feature vector of three elements. A chi-squared based cluster analysis is then used to map the reduced feature vectors to seabed type. Performance and accuracy of the approach has been established through an extensive set of trials conducted under the sponsorship of the Defence Research Establishment Pacific. A 200 kHz echo sounder operating on a tow fish was used to collect over 70 line kilometres of acoustic data from the waters around and between Vancouver and Victoria, B.C., Canada. Seabed types include mud of varying densities with and without shells, sand, gravel, cobble, and bedrock. Simultaneous side-scan imagery was collected and ground truth obtained by diver and ROV sampling. This paper describes the system architecture, implementation, and performance. The correlation between bottom types and acoustic signatures from the sea trials is discussed in a quantitative fashion. The application of these techniques to other sounders, water depths, and seabed type regimes is also discussed.