A scientific data mining approach to midwater multibeam echosounding for fisheries applications

Midwater acoustic backscatter measurements collected by multibeam sonar offer new opportunities and challenges for fisheries applications. A scientific data mining technique to handle midwater multibeam backscatter data is presented. Most of the earlier research on multibeam echosounding for fisheries has focused on the core basic technologies of multibeam transducers, the associated signal processing, and calibration. Some work has been done with post-processed data, but no systematic methodology for post-processing of midwater multibeam backscatter data has emerged. In this paper, the problem is placed in a data mining framework. A model inversion technique is utilized, by applying the inverse of an approximation to the multibeam echosounding model. The proposed approach leads to a data product consisting of a collection of midwater soundings. A multibeam midwater sounding is the equivalent of the standard multibeam soundings as obtained from hydrographic multibeam instruments. These soundings can be visualized directly, allowing for exploratory data analysis in a 3d or 4d interactive environment. A sounding is a measurement in space and time, and has associated attributes or features, such as the backscatter value. Other features can be tagged to the soundings, forming generalised data nodes. Advanced spatiotemporal data mining techniques can now be applied to this set of nodes. Some further clustering techniques are presented, clustering the soundings into groups representing coherent objects in the water column, or, more specifically, fish schools. Global properties of clusters can be derived from the individual feature tags of the soundings, thus allowing for classification of schools into classes of similar types. The latest developments of this research are presented.