Locating and surveying underwater dumping sites of unexploded ordnance (UXO) is challenging, particularly when they are buried beneath the seafloor sediment. Low-frequency (LF) sonars have a demonstrated capability for detecting buried objects. However, high clutter densities observed in relevant operational environments will limit their use unless a capacity is developed to distinguish between clutter (e.g., natural objects, debris, and seafloor topography) and UXO. Simple level-based detectors generate an impracticable number of false alarms and are therefore inadequate for this purpose. The primary objective of this project was to improve the LF sonar capability of detecting UXO by focusing on the procedure of reducing the number of contacts corresponding to clutter.

The processing chain developed in SERDP project MR-2200 was used as input to this project, together with data acquired by an experimental LF side-looking sonar in the MUD-2011 experiment funded by the Netherlands Ministry of Defense. MUD-2011 was conducted in a shallow-water estuary with a muddy bottom and the deployed objects included 155mm shells, Mk82 and Mk84 bombs, and mine-like targets; the objects were buried in approximately 0.5-1.0 m of mud.

A two-stage detection approach has been developed that uses LF Synthetic Aperture Sonar (SAS) images as input:

1. An algorithm is developed for detecting contacts that correspond to deployed objects and clutter.
2. A feature-based approach is subsequently developed to reduce the amount of contacts corresponding to clutter.

A feature-based approach to reduce the amount of clutter contacts requires a careful feature selection. Feature selection therefore received substantial effort in the current project. The robustness of the feature set and the performance of the two-stage detection approach are illustrated on data acquired in the MUD-2011 experiment. The influence of MR-2200 background suppression on the detection performance was also investigated.

The two-stage detector was trained and tested on different runs of the MUD-2011 experiment. The application of the two-stage detector approach has demonstrated its capability to reduce clutter. The figure below illustrates the result of the detection procedure on a typical LF-SAS image. All circles correspond to detections, the red ones to contacts detected as clutter, the green ones to contacts detected as UXO; white thick dashed circles denote deployed buried UXO objects, thin dotted circles correspond to deployed non-UXO objects.   

The UXO detector developed in this project will contribute towards achieving an operational capacity for performing rapid wide-area surveys of underwater military munitions sites using LF broadband sonar. High clutter densities are expected in relevant environments and, by reducing the false alarm rate, the burden on follow-on surveys and remediation efforts will be reduced, leading to more efficient and cost-effective operations. The output can also aid in localizing and sizing the sites and for performing check surveys after remediation. The results of this project also provide insights into the capabilities of LF sonar and the expected performance in operational conditions.