Multi-Sensor Physics-Based Classification of Unexploded Ordnance

Abstract

The proposed research will concentrate on how the various magnetometer and induction models Duke has developed can be used to support improved UXO discrimination, especially using multisensor data (e.g. magnetometer, frequency-domain EMI and time-domain EMI). We will utilize the multiple, offset dipolemoment model in the context of UXO classification, utilizing training data to constrain the inversion. Such constraints will be implemented as priors in a Bayesian setting. The classification algorithms will process the model parameters (features) determined by the fit to measured data. The features will be processed by several classifiers, including a likelihood ratio, a support-vector machine (SVM) and a Bayesian relevance-vector machine (RVM). The classifiers will be developed assuming knowledge of the target class (UXOs) only, assuming little or no information concerning the infinite class of false targets. Moreover, the focus will be primarily on small- and medium-sized UXO (since the large ordnance will generally be excavated in any case).

Open PDF

Document Details

Document Type
Technical Report
Publication Date
Jan 14, 2005
Accession Number
ADA433734

Entities

People

  • Lawrence Carin

Organizations

  • Duke University

Tags

Communities of Interest

  • Materials and Manufacturing Processes
  • Sensors

DTIC Thesaurus Topics

  • Algorithms
  • Classification
  • Detection
  • Detectors
  • Dipole Moments
  • Frequency
  • Frequency Domain
  • Inversion
  • Magnetometers
  • Munitions
  • Resonant Frequency
  • Signal Processing
  • Supervised Machine Learning
  • Targets
  • Time Domain
  • Training
  • Unexploded Ammunition

Readers

  • Computational Modeling and Simulation
  • Military/Explosive Ordnance Disposal (EOD) Technology
  • Neural Network Machine Learning.

Technology Areas

  • AI & ML
  • AI & ML - Bayesian Inference