The Department of Defense (DoD) is responsible for clearing many sites that are potentially contaminated with munitions as a result of past training and weapons testing activities. In many cases, these activities occurred near, or were performed in, shallow water environments where munitions pose threats to public safety and the environment. The objective of this SERDP Exploratory Development (SEED) project was to develop efficient signal processing techniques for the detection and classification of military munitions in shallow underwater environments using data collected from synthetic aperture sonar (SAS) systems.
The project team addressed the problem of detecting the presence of underwater munitions using the multichannel coherence analysis framework. The detection hypothesis was that the presence of munitions in the sonar backscatter collected from a hydrophone array would lead to higher levels of coherence compared to the backscatter from the sea floor alone. The team also worked to develop a robust target classification method that could be applied to the detected contacts to discriminate munitions from non-hostile man-made objects and competing clutter. This method was developed based on the Matched Subspace Classifier (MSC) using multidimensional Acoustic Color (AC) data extracted from the raw sonar returns. Scattering models developed by the Applied Physics Laboratory, University of Washington (APL-UW) were acquired to generate the required training dataset for various unexploded ordnance (UXO) and non-UXO objects.
The project’s first task was to develop automatic target recognition (ATR) algorithms for detection and classification of military munitions using low frequency SAS. Secondly, the project worked to develop a munition detection method using Broadband Coherence Test (BCT) that provides high probability of detection and low false alarm rate. The third task was to develop a robust munition classification system using MSC that can be trained on model-based synthetic sonar data and then be applied to real sonar datasets to successfully discriminate munitions from competing natural or man-made clutter. Lastly, the project pursued extensive testing and validation on PondEX09-10 and TREX13 data sets.
Results of this preliminary experimentation show that broadband coherence statistics are indeed capable of discerning munitions lying on the seafloor from the background. The scattering model developed by the APL-UW subcontractors allowed for generation of a large and realistic training set for a wide variety of UXO and non-UXO objects. Preliminary results of the MSC classifier trained exclusively on model-generated sonar data demonstrated good generalization ability in discriminating munitions accurately in real sonar data sets, e.g., PondEX10 and TREX13. To further validate the developed detection and classification systems, more realistic and challenging sonar data sets are needed.
Developing an effective and robust munition detection and classification system addresses the critical needs of DoD for clearing contaminated shallow underwater sites. This SEED research project identified many areas that can be pursued by future research. One of the most difficult issues for implementation is to determine a detection threshold that is robustly capable of achieving a desired false alarm rate over a wide range of environments. Although it is unrealistic to expect model data will capture all such variations for target characterization, training based on the model data will provide a classifier that is trained using a “baseline” training set.