The overall objective of this project was to investigate the use of broadband sonar to detect and classify underwater munitions near a water–sediment interface. The research combined at-sea experiments, target scattering models, and the signal processing required to test binary classification (i.e., target versus non-target). Data–model comparisons provided validation of finite-element (FE) models. Once validated, FE models are executed to obtain free-field scattering amplitudes, which are used by an acoustic ray model to investigate variations in target scattering geometry and environmental properties. The central hypothesis of this experiment was that the environment and the geometry within that environment can alter an acoustic response of a target, so the target-in-the-environment-response (TIER) must be taken into account during the development of robust detection and classification strategies.

Technical Approach

This research took a two-prong approach towards resolving issues identified in a previous effort (SERDP MR-1665) that affects sonar detection and classification algorithms of underwater UXO using sonar. Sonar data were acquired on proud, partially buried, and buried targets over a broad frequency range and aspect angle range in natural environments. These data were reduced to an inventory of TIER signatures and were used to validate the acoustic scattering models. With the TIER inventory and validated models, a study of binary classification (target versus non-target) was undertaken.

Until data from at-sea experiments became available, model validation and this project’s efforts in classification leveraged data collected during SERDP MR-1665. Under that effort, scattered acoustic signals were collected from a small set of targets in a freshwater pond with a flattened sand sediment. Some of those measurements were used to validate the FE models.

Three at-sea experiments provided additional data from an extended inventory of targets in an oceanic environment with a sand sediment and a brackish bay with a mud layer over sand. The targets included inert UXO, reference targets with well-understood TIER (e.g., a finite cylinder), and clutter items (e.g., rocks of comparable size to the UXO and scuba tanks). The first two experiments were conducted in the Gulf of Mexico during 2012 (GULFEX12) and 2013 (TREX13), and the third experiment was performed in St. Andrew's Bay (Panama City, FL) during 2014 (BAYEX14). The complexity of the environmental conditions offered by these sites spanned conditions in which UXO and munitions are found. Broadband sources and receivers, scanned along a straight rail system, were used to collect scattered acoustic signals suitable for synthetic aperture sonar (SAS) processing and generation of acoustic color templates (i.e., target strength as a function of a target-centered aspect angle and frequency).

The primary frequency band was approximately 1-30 kHz, but a higher frequency band of approximately 100-200 kHz was also used during the bay experiment. Targets were placed at horizontal ranges of 5-50 m from the Applied Physics Laboratory, University of Washington (APL-UW) rail system. Depending on sediment type and diver manipulations, the burial state of a target could be proud (i.e., on the water–sediment interface), partially buried, or fully buried. To model the interaction of sound with targets and their local environment, FE models were constructed and exercised. Comparisons of FE model predictions with at-sea data provided additional validation, and demonstrated that research carried out under SERDP MR-1665 could be transferred from a pristine pond environment to the ocean.


Acoustic scattering measurements from GULFEX12, TREX13, and BAYEX14 were used to validate new FE models for the 105-mm bullet-shaped targets, the 155-mm howitzer with and without its endcap, and a stemless scuba tank. The TIER inventory was then analyzed by a relevance vector machine (RVM) classifier for binary target classification. The sediment in GULFEX12 and TREX13 was composed of medium-fine sand with minor amounts of shell fragments. This sediment type is consistent with the sediment in PondEx10 (SERDP MR-1665). To investigate variations in the environment of TIER, BAYEX14 was conducted in St. Andrew’s Bay, FL, which was a shallow water environment (approximately 8 m depth) with brackish water and sediment consisting of a mud layer over a sand basement. The mud layer was estimated to be 15-30 cm thick.  Upon placing targets on the water–mud interface, all targets buried to some extent. Smaller, heavy targets (100-mm UXO replicas, 105-mm bullet-shaped shells) were completely buried while larger targets would be partially buried.

The TIER inventory of target signatures is currently composed of the scattered signals recorded during PondEx10 and TREX13. The Synthetic Aperture Deconvolution (SAD) algorithm was used with PondEx10 data to isolate the scattered signal for individual targets. The isolation of the scattered signals for targets deployed in TREX13 originally used the SAD algorithm. Recently, the spatial filtering step has been updated to use an algorithm based on holographic back-projection. One reason for switching from the SAD algorithm to holographic back-projection is that the SAD algorithm requires a Weiner noise parameter. This parameter is a free parameter and its value is simply set by trial and error. The database that underlies the TIER inventory is extensible. Work is ongoing to isolate the scattered signal from the BAYEX14 data. As new signals are added to the TIER inventory, the training and testing of classification schemes can be improved.

FE models have been developed for several targets under SERDP MR-2231 and an Office of Naval Research (ONR)-funded Mine Countermeasures (MCM) program at APL-UW. For the 105-mm shell and the 155-mm howitzer with an endcap, the internal material was modeled as either air or water. In all, 11 FE models were exercised to produce the free-field scattered pressure on a hemisphere centered on the target, where the source and receiver are co-located. (The radius of the hemisphere was 10 m.)  The simulated scattered pressure for each target was converted to a scattering amplitude and tabulated for use with the fast ray model. The FE+ray model allows for simulation of additional data for these targets at various ranges and different sediment types. The simulated data then can augment the experimental data in classification studies. It is important to recall that the scattering amplitude contains all the information about the target, the directionality of the scatter field, and is independent of range.  

Acoustic color templates derived from experimental and simulated data were used to test three classifiers available from the MCM community (i.e., RVM, support vector machine [SVM], and kernel matching pursuit [KMP]). These classifiers are all binary classifiers, where the outcome determines whether an object is target-like or non-target-like. The results presented in the Final Report have focused primarily on the RVM classifier. With limited experimental data, “virtual experiments” were conducted where portions of the TIER data were superposed to produce new data or augmented by simulated data. This allowed sufficient data to train and test the classifiers. It is noted that the classifiers were designed with underwater mine and mine-like objects (i.e., the targets are typically larger than those used in TREX13 and BAYEX14) in mind, and no attempts were made to tune the classifiers to UXO other than the standard training procedure.


This research provided acoustic data on a set of underwater targets under various environmental conditions. These data are the ground truth in the construction of FE models. Once validated, the FE models were exercised to determine free-field scattering amplitudes, which are required by a fast ray model. The combined FE+ray model can provide cost savings by reducing the number and/or duration of field tests. By providing a capability to simulate sonar performance, SERDP can make informed decisions on the relative merits of existing sonar systems and on proposed modifications to these systems for underwater UXO management.