Many active and former military installations have ordnance ranges/training areas with adjacent water environments in which unexploded ordnance (UXO) now exists due to wartime activities, dumping, and accidents. Innovative technologies able to separate UXO from false targets and to discriminate amongst UXO targets themselves are needed. The objective of this project was to address the scientific and technical issues whose resolution would result in an efficient, high performance structural acoustic (SA) feature-based underwater sonar technology able to detect and localize buried (and proud) targets and separate the detections into UXO vs. non-UXO. The focus was on marine-based sonars that could look both downward and sideways in water depths ranging from several meters to tens of meters. The goal is to develop a sonar approach that results in robust identification algorithms based on SA features and complementary 3-D synthetic aperture sonar (SAS) images and to demonstrate the ability to detect and classify proud and buried UXO in the presence of natural and man-made clutter with actual SA sonar systems at sea.
In the SA regime, acoustic wavelengths are comparable to the target dimensions, and sound readily penetrates the target and the sediment with the echoes directly related to the vibrational dynamics of the target. Aspect-frequency features can then be used to “fingerprint” the target without requiring a high resolution image. This project developed approaches allowing the sonar to capture echo data over various aspects and angular apertures to provide robust target echo spatial/spectral maps (acoustic color) and the associated classifying features. Based on the observed scattering and the operative SA mechanisms, various “fingerprints” were explored, as well as how they should be extracted from the measured scattering patterns. Practical experimental/numerical methods were developed for training the relevance vector machine (RVM) identification algorithms and demonstrating the advanced classifying sonar technology in a marine environment. This requires the development and demonstration of advanced statistical classifier techniques that have the ability to mitigate adverse effects on the identification algorithms caused by target burial in an absorptive sediment, unknown vertical orientations, UXO target variations, and variability in the acoustic environment and clutter. The project team envisioned a SA sonar implemented on an underwater platform/vehicle utilizing broadband, low frequency, compact acoustic source and receiver technology. Scattering data was post processed in a synthetic aperture fashion providing multi-static scattering data from small volumes of the sediment at various target aspects for submission to the SA feature-based identification algorithm. After demonstration of the sonar technology and associated SA identification algorithms, existing sonars, or ones designed according to the project findings and acquired by an industrial organization, could then be utilized in actual clean-up operations.
This project demonstrated the structural acoustic technology with an autonomous underwater vehicle (AUV)-based SA sonar successfully detecting UXO buried in the sediment in the Gulf of Mexico and showing that their SA features could be used to separate them from proud and buried false targets. The Buried Object Scanning Sonar (BOSS) was set up to fly at several meters altitudes in 18.3 meter water depths off the coast of Panama City in the Gulf of Mexico. This AUV-based sonar carried out north-south, east-west, and diagonal paths over a target field comprised of nine explosive simulant-filled UXOs buried about 10cm below the sediment/water interface, two buried false targets, and 23 proud targets. The scattered acoustic pressure signals from the target field received at each of the 40 wing sensors as the vehicle moved in a straight line were processed in a synthetic aperture manner yielding both 3-D images and several acoustic color constructs for all the buried targets and for seven of the proud targets considered to be non-UXO. Most of the images gave useful information related to the target size and burial orientation. A multi-dimensional feature extracted from the acoustic color maps demonstrated almost perfect separation between the nine UXO and the nine false targets. These results demonstrate that typical buried UXO can be detected, imaged, and classified (UXO versus non-UXO) using a SA sonar and a relevance vector machine identification algorithm. In addition, earlier measurements made with a rail-based SA sonar in 25 foot waters off the Duck, NC, coast validated the new concept of short time (specular) versus long time (elastic highlight) plan view imaging. Early in the project, the first accomplishment involved the sediment pool demonstration of a numerically trained classifier. Here the project team demonstrated good classifier performance training the relevance vector machine algorithm on a finite element target burial angle simulation data set and testing on UXO and false targets buried in the sediment pool.
The high resolution imaging sonar technologies available today can prosecute targets lying on the sediment surface, but their high acoustic frequencies prevent seeing UXO buried in the sediment. Several SERDP projects (MR-1513, MR-2103, etc.) have been exploring SA-based sonar for detection and classification of underwater UXO. The structural acoustic approach offers significant advantages over more conventional acoustic approaches, which rely only on the formation of high resolution images. These advantages include a diverse set of spatial and spectral SA “fingerprints” leading to high probability of detection, low false alarm rates, and low frequency sediment penetration permitting buried target prosecution. Further, the SA approach allows the formation through SAS processing of complementary 3-D images of the sediment volume and of any targets buried therein. Even though the SA frequencies are relatively low, these images have sufficient resolution to allow determination of the approximate target size, burial depth, and burial angle. The combination of this information together with target identification/classification through SA “fingerprints” provides the necessary information regarding the presence, location, and identification of underwater UXO for effective inspection at sites requiring remedial action. Furthermore, the ability to use a high fidelity finite-element-based numerical model (addressing both the complex elastic target and the acoustic propagation environment) to generate simulated data for classifier training is extremely important given the many targets, burial conditions, and environments of interest.