Use of Target Shape and Size in Classification of UXO in Survey Data
Abstract
AETC, working with our subcontractor and partner, Visual Learning Systems, Inc. (VLS) has completed Year 1 of our development of a UXO target picker and probabilistic classifier based primarily on shape metrics within an inductive learning software environment for detecting and discriminating UXO. This effort builds upon our prior research effort in the Seed Project, SERDP UX-1354. In the Seed Project, working with vehicular towed-array data from the 2001 Badlands Bombing Range (BBR) UXO survey, we showed that: Little or no further performance gain can likely be achieved using machine-learning techniques that focus only on the physics-based fitting parameters to make decisions about ordnance classification; There is substantial value in the use of shape and size information derived from target images created from high-density mapped sensor data to make ordnance classification decisions; Pattern recognition techniques that incorporate size and shape information can dramatically reduce the number of non-UXO targets while maintaining most of the sensitivity of traditional physics-based approaches for finding true UXO.
Document Details
- Document Type
- Technical Report
- Publication Date
- Dec 01, 2005
- Accession Number
- ADA552614
Entities
People
- David W. Opitz
- Jim R. McDonald
- Stuart Blundell