Use this URL to cite or link to this record in EThOS:
Title: Learning from sonar data for the classification of underwater seabeds
Author: Atallah, Louis N.
ISNI:       0000 0001 3431 4604
Awarding Body: University of Oxford
Current Institution: University of Oxford
Date of Award: 2005
Availability of Full Text:
Access from EThOS:
Full text unavailable from EThOS. Please try the link below.
Access from Institution:
The increased use of sonar surveys for both industrial and leisure activities has motivated the research for cost effective, automated processed for seabed classification. Seabed classification is essential for many fields including dredging, environmental studies, fisheries research, pipeline and cable route surveys, marine archaeology and automated underwater vehicles. The advancement in both sonar technology and sonar data storage has led to large quantities of sonar data being collected per survey. The challenge, however, is to derive relevant features that can summarise these large amounts of data and provide discrimination between several seabed types present in each survey. The main aim of this work is to classify sidescan bathymetric datasets. However, in most sidescan bathymetric surveys, only a few ground-truthed areas (if any) are available. Since sidescan ‘ground-truthed’ areas were also provided for this work, they were used to test feature extraction, selection and classification algorithms. Backscattering amplitude, after using bathymetric data to correct for variations, did not provide enough discrimination between sediment classes in this work which lead to the investigation of other features. The variation of backscattering amplitude at different scales corresponds to variations in both micro bathymetry and large scale bathymetry. A method that can derive multiscale features from signals was needed, and the wavelet method proved to be an efficient method of doing so. Wavelets are used for feature extraction in 1D sidescan bathymetry survey data and both the feature selection and classification stages are automated. The method is tested on areas of known types and in general, the features show good correlation with sediment types in both types of survey. The main disadvantage of this method, however, is that signal futures are calculated per swathe (or received signal). Thus, sediment boundaries within the same swathe are not detected. To solve this problem, information present in consecutive pings of data can be used, leading to 2-D feature extraction. Several textural classification methods are investigated for the segmentation of sidescan sonar images. The method includes 2D wavelets and Gabor filters. Effects of filter orientation filter scale and window size are observed in both cases, and validated on given sonar images. For sidescan bathymetric datasets, a novel method of classification using both sidescan images and depth maps is investigated. Backscattering amplitude and bathymetry images are both used for feature extraction. Features include amplitude-dependent features, textural features and bathymetric variation features. The method makes use of grab samples available in given areas of the survey for training the classifiers. Alternatively, clustering techniques are used to group the data. The results of applying the method on sidescan bathymetric surveys correlate with the grab samples available as well as the user-classified areas. An automatic method for sidescan bathymetric classification offers a cost effective approach to classify large areas of seabed with a fewer number of grab samples. This work sheds light on areas of feature extraction, selection and classification of sonar data.
Supervisor: Probert Smith, Penny Sponsor: Karim Rida Said Foundation
Qualification Name: Thesis (Ph.D.) Qualification Level: Doctoral
EThOS ID:  DOI: Not available
Keywords: Sensors ; Ocean and coastal engineering ; sensors ; sonar ; machine learning ; classification ; acoustics ; underwater