Using wavelet analysis to classify and segment sonar signals scattered from underwater sea beds

Journal article

This work is concerned with the automatic characterisation and classification of seabed sediments by using wavelet transform techniques to analyse the incoming one-dimensional signals from both sidescan and sidescan bathymetric sonars. This method extracts features from the energies at different scales of the wavelet transform of the signal then uses these features to classify different types of sediments. The features selected include the sum and standard deviation of the wavelet coefficient energies. These features are then given to a neural network for classification, and classification results are compared. Three datasets were provided, one sidescan sonar data set and two sidescan bathymetric sonar datasets. The sidescan dataset was already corrected, but the signals from the sidescan-bathymetric dataset were corrected for losses. The method is also tried on the same sediment type from two different datasets. Compared to only using properties from the power spectrum to classify sediments, the method provides the user with an efficient tool to observe features of sediments in both time and scale. It is a fast method that can be applied online. The rates of correct classification using the features as inputs to an MLP neural network were more than 98% when applied to the sidescan dataset.

Authors: Atallah, L; Smith, P

• Journal: Acta Acustica united with Acustica
• Volume: 88
• Issue: 5
• Pages: 615-618
• Publication date: 2002-09-01
• ISSN: 1436-7947
• Language: English