Hidden Markov Models for Sensor Fusion of EMI and GRP

Abstract

An investigation of hidden Markov models (HMMs) as a means of processing Ground Penetrating Radar signals for discrimination between anti-tank (AT) and anti-personnel (A')) landmines and discrete clutter objects. Hidden Markov models had been used successfully previously in AT landmine detection with GPR but had not been used as a discriminant between mines and clutter nor had they been used on AP mines. Experiments were conducted on data collected at the JUXOCO calibration grid. Continuous and discrete HMMs were trained and tested and evaluated on the grid using both the Baum-Welch and discriminative training algorithms. Experimental results suggest that discriminative training algorithms should be used for training HMMs for landmine detection, fusion of continuous and discrete models provided improved performance, and that, although HMMs with the existing feature sets perform well for detection of AT mines, new feature sets should be developed for discriminating between mines and clutter objects.

Open PDF

Document Details

Document Type
Technical Report
Publication Date
Nov 17, 2003
Accession Number
ADA422405

Entities

People

  • Paul Gader

Organizations

  • University of Missouri

Tags

Communities of Interest

  • Counter IED
  • Energy and Power Technologies
  • Human Systems
  • Sensors

DTIC Thesaurus Topics

  • Algorithms
  • Calibration
  • Detection
  • Detectors
  • False Alarms
  • Ground Penetrating Radar
  • Hidden Markov Models
  • Information Science
  • Land Mines
  • Markov Models
  • Models
  • Probability
  • Probability Distributions
  • Radar
  • Sensor Fusion
  • Stochastic Processes
  • Training

Readers

  • Military/Explosive Ordnance Disposal (EOD) Technology
  • Sensor Fusion and Tracking Systems.
  • Speech Processing/Speech Recognition.