Many submerged unexploded ordnance (UXO) are buried under seafloor sediment, making them hard to detect. Though buried, they can resurface after storms or be encountered during dredging. This project aims to enhance UXO detection and classification capabilities of sediment penetrating low-frequency acoustic imaging systems such as the Buried Object Scanning Sonar (BOSS). This will be accomplished by developing methods for extracting high signal-to-noise ratio impulse response measurements of targets from beamformed data, developing data-driven methods for reducing 3D SAS imaging distortion, and providing operators with improved visualization methods for interpreting BOSS data.

This project will simulate time-series data collected by a downward looking low frequency synthetic aperture sonar (SAS) system. The simulated data will be used to develop an efficient beamformer that extends fast-factorized-back-projection to three dimensions. In combination with simulations, the beamformer will be used to test data-driven autofocus techniques adapted from previous work done in conventional low frequency SAS. Target impulse response functions will be extracted from BOSS imagery by extending techniques that are currently used for low frequency, side-looking SAS data to the 3D, downward looking geometry of the BOSS systems. Spatial coherence estimation methods will be tested for optimizing the spatial masks used during signal extraction. Coherence, spectrum and aspect information will be encoded into the volume representations of 3D SAS images using voxel color and transparency. Following the initial development and testing of signal extraction and beamforming methods on simulated data, the algorithms will be tested on data captured by fielded BOSS systems.

This project will benefit DoD by improving probability of detection and false alarm rates for downward looking, sediment penetrating synthetic aperture sonar systems. One of the main difficulties associated with exploiting structural acoustic information from buried targets is the low signal-to-noise ratio that is commonly encountered. This project aims to develop methods for improving the signal-to-noise ratio of structural acoustic response measurements and provide a signal-processing framework for integrating classifiers exploiting structural acoustics into the BOSS signal processing chain. This will improve classification performance and allow the relative effectiveness of classifiers to be compared. Additionally, systems such as BOSS that exploit aperture synthesis to improve resolution tend to produce degraded imagery when navigation information is insufficiently accurate. This project will develop methods for using information from the backscattered acoustic signals to augment external navigation sources and generate optimally focused 3D SAS imagery, improving shape-based classification performance and the imagery available to operators.