Abstract
AbstractThe detection of and response to underwater munitions will undoubtedly require the appropriate combinations of fully integrated sensors and imaging systems and platforms, as well as navigation and positioning technologies, to handle the variability in bottom conditions, water clarity and depth, size and type of munitions of interest, whether they are buried or proud. Where visibility allows, practically no sensing modality matches the details and information content from optical imaging systems for target localization, discrimination and identification. The significant disadvantage of optical systems for underwater applications is the range limitation. Sonar imaging systems are of limited resolution but do not have such a severe range limitation, as acoustic energy propagates well through turbid waters.In this study, we have explored two aspects of the munitions detection and classification process: (1) high-resolution mapping of an environment using a high-frequency sonar system to determine footprints of areas with munitions present and target localization in a wide-area survey and to perform detailed surveys for individual detected items during a re-acquisition process and (2) Multiple-Aspect Fixed-Range Template Matching (MAFR-TM) for detection and classification of the potential target.The MAFR-TM approach was tested using (1) a singular target scene collected in a test tank, (2) a cluttered scene acquired in the same test tank, and (3) a cluttered scene obtained in a realistic field environment (a marina). The munitions-like targets were cylinders made of steel or aluminum. The clutter was a collection of PVC tubes. Biological growth surrounded the target and artificial clutter in the marina. The experimental results indicate that the detection algorithm performs fairly well with the tank data (100% of the targets are detected) and cluttered tank data (94.44%). The classification between metals and plastics, proper orientation and target localization is also of good quality: 94.4% of the detected targets are properly classified as metal alloy if no clutter is present versus 82.35% in the presence of clutter. The algorithm performance in the marina is reasonably good, even though the overall performance drops: 61.11% of the targets are detected, and 68.18% of the detected targets are properly classified as metal alloy.