Abstract

Seafloor segmentation and characterization based on local textural properties of acoustic backscatter has been a subject of research since 1980s due to the highly textured appearance of sonar images. The approach consists of subdivision of sonar image in a set of patches of certain size and calculation of a vector of features reflecting the patch texture. Advance of multibeam echosounders (MBES) allowed application of texture-based techniques to real geographical space, and predicted boundaries between acoustic facies became experimentally verifiable. However, acoustic return from uncalibrated MBES produces artifacts in backscatter mosaics, which in turn affects accuracy of delineation. Development of Geocoder allowed creation of more visually consistent images, and reduced the number of factors influencing mosaic creation. It is intuitively clear that more accurate backscatter mosaics lead to more reliable classification results. However, this statement has never been thoroughly verified. It has not been investigated which corrections are important for texture-based characterization and which are not essential. In this paper the authors are investigating the Stanton Banks common dataset. Raw data files from the dataset have been processed by the Geocoder at different levels of corrections. Each processing resulted in a backscatter mosaic demonstrating artifacts of different levels of severity. Mosaics then underwent textural analysis and unsupervised classification using Matlab package SonarClass. Results of seafloor characterization corresponding to varying levels of corrections were finally compared to the one generated by the best possible mosaic (the one embodying all the available corrections), providing an indicator of classification accuracy and giving guidance about which mosaic corrections are crucial for acoustic classification and which could be safely ignored.

Department

Center for Coastal and Ocean Mapping

Publication Date

7-2012

Journal Title

European Conference on Underwater Acoustics

Conference Date

July 2 - July 6, 2012

Publisher Place

Edinburgh, UK

Document Type

Conference Proceeding

Share

COinS