Characterisation and classification of the sea-floor from the power-spectra of side-scan sonar traces

This work is concerned with the automatic characterisation and classification of the sea-bed for side-scan sonar trace power-spectra.A parametric model of side-scan sonar trace power-spectra is developed from the equation for the magnitude frequency response of a Butterworth filter. The model's parameters are optimised to give a least squares fit with observed spectra. Three of the optimised parameters are used to define features.The parametric optimisation approach to feature extraction is compared to a method described by Pace and Gao (1988) in which features are defined in terms of ratios of integrals over frequency intervals of observed (specially defined) power spectra.Scatter distributions in feature space are reduced to sets of numbers that define distribution fields and these constitute seabed characteristics. A classification exercise is undertaken to demonstrate the utility of the feature extraction methods. The discrimination between sea-bed types achieved by the feature extraction methods is succintly conveyed in class discrimination matrices.The merits of the spectral modelling approach to feature extraction are: (1) the features provide a meaningful description of the spectra from which they are extracted, (2) the features enjoy a degree of immunity to changes in noise in the signal from which they are extracted, and (3) the features provide excellent discriminants. The parametric fitting process, however, is slow. An important merit of the Pace and Gao (1988) approach is that feature extraction is rapid.

Sea-bed characterisation and classification from the power spectra of side-scan sonar data. Available from: http://www.researchgate.net/publication/226366553_Sea-bed_characterisation_and_classification_from_the_power_spectra_of_side-scan_sonar_data [accessed Dec 7, 2015].