Description
The aim of our study was to show that acoustic data are suitable for seabed mapping. We developed a methodology, which includes all the necessary steps from data acquisition to seabed classification.
Mapping of the seabed can be done visually, mechanically or acoustically. All visual methods (divers, video recording, photography) and mechanical methods (in-situ sampling) are slow and require a lot of effort, and consequently they are expensive and unsuitable for mapping large areas of the seabed. Additionally, usefulness of optical and laser methods is limited in the Slovenian sea and generally in a large part of the northern Adriatic due to the very turbid water.
Multibeam sonars use audio signals, which are independent of the transparency of the water and allow us to collect high quality data even in very turbid water. Multibeam sonar provides continuous coverage along with high speed of acquisition. We used the Reson Seabat 8125 multibeam sonar for data acquisition. We obtained two types of data – bathymetries and acoustic intensities. Both types of raw data from the multibeam sonar were processed in our data processing module. The data processing module included verification of data quality, elimination of the impact of system settings, normalization for incidence angle, derivation of features and seabed classification using data mining. Verification of data quality was based on a median filter for image processing. After the verification of data quality, elimination of the impact of system settings and normalization for incidence angle was performed by means of empirically derived formulas. The interaction between the acoustic signal, sea-water and seabed under different incidence angles is too complex for theoretical treatment, which is why an empirical approach using planned experimental measurements and comparison of results was taken. Further, we derived features from pre-processed data. We divided both types of data – bathymetries and intensities of acoustic reflections – into patches sized from 1x1 meter to 4x4 meters. For every patch specific simple features like average value, standard deviation and partial derivatives were computed and more complex texture features like higher moments and grey-level co-occurrence matrix properties were derived. All features were then evaluated, the most informative features were selected and used in machine-learning algorithms to produce seabed maps.
Based on the developed methodology, measurements carried out, acoustic data processed and areas of seabed types were determined for chosen area of the Slovenian sea. We confirmed the effectiveness of our method and demonstrated that seabed classification using multibeam sonar data has great potential in seabed research. Our method enables fast and efficient mapping of large areas of seabed and provides continuous coverage throughout the region.
Primary author
Mr
Sašo Moškon
(Harpha Sea, d.o.o. Koper)
