Large-scale Seabed Classification: from Image Classification to Seabed Geology
Xavier Monteys Geological Survey of Ireland; firstname.lastname@example.org
High frequency multibeam (EM120 & EM1002) bathymetry and backscatter datasets acquired during the Irish National Seabed Survey (2000-2005; www.infomar.ie), in water-depths ranging between 100- 5,000m, have been used in this study to characterise over 700,000 km² of the Irish seabed (ca. 7 times the size of Ireland). Large-scale automated image classification on seabed’s backscatter data has been performed, to provide baseline physical seafloor maps and sediment type databases, primarily to be utilized in marine habitat mapping and large-scale environmental studies (e.g. IOSEA). In general, the multibeam backscatter response has allowed us determining a number of consistent and statistically distinct classes throughout the study area, providing a continuous proxy for seabed geological properties. In deep-waters these classes are closely related to soft-hard and smooth-rough descriptors and the primary control over the backscatter classes are intensity features (e.g. mean, quantiles). However, in water depths between 100-500m, spectral image characteristics (particularly variance and GLCMs) provide valuable insight into the morphological aspects of the near-seabed composition, particularly in coarse-grained sediment’s seabed types. On the other hand, Bathymetric features, such as slope and rugosity, have been derived, integrated and correlated to the backscatter classification results providing good results for rock outcrop characterization. Furthermore, classification scales & resolution, image features suitability, ground-truthing validation and interpolation techniques have been topics examined and discussed in a large-scale context.
Access presentation here (10 meg)
Return to Presentation Programme Page