Examining statistical segmentation of multibeam backscatter images with Geovisual Analytics
Kazi Ishtiak Ahmed1, Urška Demšar1, Xavier Monteys2
1National Centre for Geocomputation, National University of Ireland Maynooth; email@example.com 2 Geological Survey of Ireland
Multi-Beam Echo Sounders are often used for classification of seabed type, as there exists a strong link between sonar backscatter and sediment characteristics of the seabed. There are a number of automated classification methodologies, most of them using image segmentation techniques. One of the procedures analyzed uses sets of statistical features extracted from the backscatter image, applies Principal Component Analysis to reduce dimensionality and k-means clustering to assign classes. This paper examines the complexity of this particular method using a Geovisual Analytical approach by exploring the features with a Self-Organizing Map (SOM). This research is a work in progress and the paper presents only the first preliminary experiment of using a SOM for this problem. The ultimate goal of the project is to examine the potential of Geovisual Analytics to help reduce the complexity of the MBES image classification methods and thereby facilitate seabed type characterization.
Access presentation here (2 meg)
Return to Presentation Programme Page