Mission-adaptive computer aided detection (CAD) for mine hunting: system concept and theory

PDF

Authors
  1. Kessell, R.T.
Corporate Authors
Defence Research Establishment Atlantic, Dartmouth NS (CAN)
Abstract
CAD systems require detection and rejection training--the adjustment of internal coefficients to discriminate between objects as required. Ideally this machine training would be carried out once for all missions by pattern recognition experts. But the range of mission parameters that affect CAD training is very large, making a once-for-all approach particularly difficult, if not impossible. Final on-site mission adjustments are inevitable. And as CAD pattern recognition schemes become more advanced, these on-site adjustments are likely to become more complex. One hurdle to significantly improved CAD, then, is the added burden that more advanced pattern recognition schemes place on the sonar operator, to whom it falls to make the final system adjustments, by manually entering parameters that presume an ever greater working knowledge of image processing. This paper presents the concept of automatic and non-intrusive on-site training for CAD, in chich the system gathers its training about targets and clutter at the start of a mission, simply by "eavesdropping" on the sonar operator's routine responses to the sonar image (selecting potential threats and passing over clutter as ordinarily expected of the operator). This automatic approach makes it possible to use more advanced pattern recognition, achieving better CAD performance without sacrificing mission flexibility or taxing the sonar operator.

Il y a un résumé en français ici.

Keywords
Computer aided detection;Target identification;Computer aided classification;Classification algorithms;ATR (Automatic Target Recognition);Performance specification
Report Number
DREA-TM-2000-111 — Technical Memorandum
Date of publication
01 Feb 2001
Number of Pages
30
DSTKIM No
CA020120
CANDIS No
516765
Format(s):
Hardcopy

Permanent link

Document 1 of 1

Date modified: