Difficulties arise in effective detection and identification of unexploded ordnance (UXO) within clutter-filled underwater environments, mainly due to imperfections in the gathered data (i.e. missing, incomplete, uncertain and/or ambiguous entries). No single sensing technology can be both accurate and cost-effective for estimating the density and spatial distribution of UXO objects, let alone identifying them. An effective mechanism is required to enhance the discrimination capability by making use of various heterogeneous data from different sensing modalities and resolutions.
The objectives of this project were to explore optical and acoustic monocular and stereo imaging for the detection and discrimination of underwater munitions and to explore a framework for fusing heterogeneous data that is suited to the UXO identification problem.
Where visibility allows, practically no sensing modality can match the information content from optical image systems, due to high resolution and data rate, as well as the rich visual cues. Detection and classification capabilities can be further enhanced by making use of a large-area visual map by mosaicing a large number of images each covering only a small field of view. In addition, sonar systems will likely play a critical role in UXO detection because acoustic energy can penetrate through silt, mud and various sources of turbidity. Finally, a suitable framework is necessary to allow the fusing of heterogeneous raw and processed data to achieve UXO identification effectively.
Acquiring data under identical imaging conditions allowed the researchers to perform a direct comparison between two existing optical cameras, DIDSON and BlueView systems. DIDSON provided better target details due to the higher spatial resolution associated with narrower horizontal beam widths.
The researchers determined that high-frequency 2-D FS sonar systems can provide and effective technology for UXO detection, and enhance the classification capability. Further research is necessary to develop robust algorithms to experiment with a larger database of targets and environmental conditions.
The researchers concluded that the ambiguity and imperfection handling mechanisms intrinsic to the data fusion method based on the Dempster-Shafer (DS) belief theoretic framework are better suited in the underwater environment. This is mainly because the DS theoretic framework requires little or no recourse to the types of assumptions that encumber traditional approaches.
This project lays the preliminary groundwork to address how streaming data generated from sensors could be utilized in real-time for UXO object identification.