Textural Analysis and Statistical Investigation of Patterns in Synthetic Aperture Sonar Images
Abstract
Textural analysis and statistical investigation of patterns in synthetic aperture sonar (SAS) images is useful for oceanographic purposes such as biological habitat mapping or bottom type identification for offshore construction. Seafloor classification also has many tactical benefits for the U.S. Navy in terms of mine identification and undersea warfare. Common methods of texture analysis rely on statistical moments of image intensity, or more generally, the probability density function of the scene. One of the most common techniques uses Haralick's Grey Level Co-occurrence Matrix (GLCM) to calculate image features used in the applications listed above. Although widely used, seafloor classification and segmentation are difficult using Haralick features. Typically, these features are calculated at a single scale. Improvements based on the understanding that patterns are multiscale was compared with this baseline, with a goal of improving seafloor classification. Synthetic aperture sonar (SAS) data was provided by the Norwegian Research Defense Establishment for this work, and was labeled into six distinct seafloor classes, with 757 total examples. We analyze the feature importance determined by neighborhood component analysis as a function of scale and direction to determine which spatial scale and azimuthal direction is most informative for good classification performance.
Document Details
- Document Type
- Technical Report
- Publication Date
- Sep 01, 2022
- Accession Number
- AD1200604
Entities
People
- Suluck P. Klumprasert
Organizations
- Naval Postgraduate School