Underwater sites affected by unexploded ordnance (UXO) pose significant risks to both human safety and environmental well-being. Sonar imaging is commonly employed to investigate such sites and aid in UXO remediation efforts. However, manually identifying and classifying potential targets in sonar data is challenging and time-consuming. Automated methods that can detect and accurately classify UXO objects is therefore desired and has been a topic of significant recent investigation. Existing approaches broadly encompass two classes of methods: 1.) physical model based techniques that utilize the knowledge of underlying phenomenology to build pre-defined detection and classification rules, and 2.) machine learning (ML) techniques that learn both features of interest and corresponding decision rules using example labeled training SONAR image samples. Most existing ML methods however are black-box, lack transparency and fail to consider the underlying physical acoustics. Additionally, acquiring sufficient training data for these ML algorithms can often be problematic. In this study, a novel approach that combines the merits of physical model based methods along with state of the art ML viz. deep learning techniques is presented.

This project designed neural networks that explicitly account for the unique physics involved in the problem domain. UXOs examined using low-frequency sound frequently exhibit resonant behavior, where the sound is re-radiated after initial geometric scattering, owing to the elastic and compressional properties of the objects. Moreover, such resonant effects are typically absent in clutter objects, making them advantageous in discriminating UXOs from non-UXOs. Consequently, the project team suggests several neural network architectures that leverage these resonant effects, utilizing three-dimensional (3D) data obtained from a synthetic aperture sonar (SAS) imaging sonar. The project team incorporates a recurrent neural network to model the physics-based correlation among adjacent time/spatial slices, originating from the resonant phenomena. The project team employs intensity imagery of orthogonal projections of the 3D data cube, which capture shape-specific resonant scattering mechanisms unique to specific types of UXOs. To evaluate the effectiveness of the methods, the project team compared them against recent state-of-the-art algorithms using a real-world 3D SAS dataset. Remarkably, even when confronted with limited training data, the approaches consistently demonstrated superior results.

The findings highlight the significant potential of incorporating physical acoustics into neural network designs for UXO detection and classification, offering improved accuracy and efficiency in underwater remediation operations.

The follow-on work recognizes that high quality image formation and a true 3D characterization of resonant scattering profiles corresponding to key, canonical target types will be crucial to success that can be translated to the field. The project team intends to explore new phenomenology enriched networks that can benefit from prior models of 3D resonant scattering of distinct UXO and other objects. The project team also plans to expand the investigation beyond the Sediment Volume Search Sonar dataset by partnering with other performers (e.g. the MUST imaging sensor from University of Washington) on the SERDP Munitions Response program, so the versatility of the project can be comprehensively demonstrated.