Benthic habitat mapping by autonomous underwater vehicles

This work presents a functional system for performing unsupervised classification and mapping of benthic habitats using an autonomous underwater vehicle. Traditionally, tne-scale underwater mapping has been an expensive process, inaccurate, and is performed infrequently. This work provides contributions in three main areas: the control systems allowing the underwater vehicle to perform autonomous measurements using inexpensive sensors; implementation of a memetic algorithm for unmixing hyper-spectral signals to identify and classify habitat types; detection of new end-member types, and the production of accurate maps, while mitigating for constraints in vehicle capabilities. Meta-data is recorded and associated with detected end-member types to act as an aid in expert classification. The production of multi-layer maps is demonstrated. These maps account for uncertainty in the vehicle's position, and variability in unmixing accuracy and confidence in data quality. This system produces end-member library spectra, maps showing end-member location and abundance, bathymetry, coverage maps, and a confidence map. An analysis of the accuracy of the algorithms using image- processing techniques showing the strength of the mapping algorithms in the presence of noise is presented. This work has developed techniques to support fine-scale benthic classification using underwater vehicles. This capability can complement existing remote sensing techniques, allowing mapping where benthic habitat is obscured from airborne and satellite sensors because of water clarity and depth. The techniques for accurate and fine-scale measurement of benthic mixtures proposed here offer alternative tools for the monitoring and management of estuarine environments.