UACE2017 Proceedings: UXO identification using a matrix pencil approach

• Session:
  Underwater Unexploded Ordnance (UXO) Detection and Remediation
• Paper:
  UXO identification using a matrix pencil approach
• Author(s):
  Mohannad Shehadeh, Raviraj Adve, Vincent Myers
• Abstract:
  The detection and identification of Unexploded Ordnance (UXO) at former military sites using sonar presents significant challenges which can be attributed to the presence of clutter objects of similar size and shape to targets of interest. Remediation of large UXO sites requires computing features that provide Automatic Target Recognition (ATR) algorithms with the ability to reliably discriminate between man-made and clutter objects based on their acoustic returns. This paper presents the use of the Matrix Pencil (MP) algorithm to enable the identification of underwater targets potentially observed from multiple aspect angles. The MP algorithm models the signal reflected from a target as linear combination of exponential modes. Importantly, using the MP approach requires only a single data snapshot to estimate the modes, i.e. we do not require an estimate of any second order statistics. The modes are determined as the generalized eigenvalues of a pair of matrices created using a single-time sample shift operator. The modes estimate the resonant frequencies (the imaginary component) and the attenuation (the real component) of the ensonified objects. The amplitudes of the modescan can then be obtained using a least-squares estimate. The Matrix Pencil approach is tested using the well-known PONDEX dataset, which consists of calibrated scattering data from different aspects of several targets, including UXO shells, collected using a rail system. Results show that the MP approach, using just the real component, is able to discriminate between classes of targets, specifically hollow versus filled shells. Intuitively, a hollow target shows lower attenuation factors; a characteristic the MP algorithm is able to quantify. Further analysis combining both the real and imaginary components could lead to even more robust identification of UXO objects.
•   Download the full paper

Contact details

• Contact person:
  Mr Vincent Myers
• e-mail:
  
• Affiliation:
  Defence R&D Canada
• Country:
  Canada