Buried underwater unexploded ordnance (UXO) resides in a challenging region of the acoustic sensing trade space. Downward-looking high frequency (≥100 kHz) real aperture sonar systems can provide excellent, high resolution maps of the sediment surface, but attenuate strongly with depth, and fail to provide definitive characterization of buried targets as quantified, e.g., by high probability of detection (P_D) and low probability of false alarm (P_FA). Low frequencies (≤ 10 kHz) penetrate the sediment, but provide much poorer spatial resolution—a few "pixels" on a UXO-sized target, and even then, only if a sufficiently large receiver array (≥ 2 m diameter) is available. 

This research was motivated by past work on electromagnetic induction (EMI) detection and discrimination of ground-based buried UXO, where metal targets are easy to detect and localize, but much more difficult to identify. The objective of this project was to identify additional spectral signatures in relatively low resolution images of buried underwater targets that robustly distinguish UXO from similarly sized clutter.

The holographic technique takes advantage of the fact that hollow metal shells exhibit multiple, distinctive acoustic resonances in precisely the 3-10 kHz band of interest for sediment penetration. Similarly sized solid objects, such as rocks, also exhibit resonances, but generally above 10 kHz. Resonances of a shell surrounded by sediment (or water) are much less sharp than those of a shell in air (quality factor Q values ~ 5 vs. ~ 10²) due to the lower acoustic contrast; however, the models show that highly distinctive spectral features still remain, and this is borne out by existing measurements.

This project sought to develop a model that is as faithful as possible to the physics, but numerically efficient. This is accomplished by restricting attention to spherical shell targets, for which full exact analytic solutions exist. Although missing some details specific to more elongated UXO geometries, spherical shells capture the essential physics of acoustic resonance, target environment interactions, and provide complete, realistic sonar responses in a computationally efficient manner.

The acoustic modeling showed that strong resonant features survive the coupling of the shell to the sediment or water environment. The exact solution for spherical targets exists only when it is embedded in an infinite homogeneous background medium. Approximations are required to treat a target buried in sediment, which essentially involves ignoring multiple scattering between the target and the sediment surface. 

The project team investigated the signal processing chain, in which the real-plus-synthetic array data is used to form 2+1D space-frequency images back-projected to a given fixed depth. The spatial image localizes targets at the 10 cm horizontal resolution level. Although detailed target features (e.g., tail fins) are not visible on this scale, the frequency dimension clearly displays the spectral fingerprint. The results show that, although not directly inferred, the true target depth may be estimated by optimizing over the assumed depth using an image focus metric.

Simulation studies were used to validate the approach for UXO localization and discrimination. The project team demonstrated the automated resonance feature extraction algorithm’s strong false alarm rejection capability.

In designing the physics-based model and holographic image-based target detection and discrimination algorithm, this project captured the key physical features of the sensor parameters, measurement grid geometry, water and sediment acoustic propagation characteristics, and target elastic response. Certain elements can be captured only qualitatively (e.g., realistic UXO resonant responses), but given the correctness of the basic physics, this will not affect the key conclusions of the analysis.

By focusing on the 3-10 kHz sediment-penetrating frequency range, it is expected that buried UXO-sized shells will exhibit sharp resonant features, with Q ≥ 10, in strong contrast to typical clutter responses with Q ≤ 1.