RATE-CONSTRAINED TARGET DETECTION. SUMMARY REPORT

PDF

Authors
  1. Goulding, M.M.
  2. Bird, J.S.
Corporate Authors
Simon Fraser Univ, Burnaby BC (CAN) Underwater Research Lab;Defence Research Establishment Ottawa, Ottawa ONT (CAN)
Abstract
Real-world detection systems, such as radars, control resources that are limited in their processing ability. This report introduces a new detection criterion designed to operate under the limitations imposed by such resource constraints. In contrast to the Neyman-Pearson framework which constrains the false-alarm rate, the rate-constraint criterion maximizes the detection probability of the test subject to a constraint on the threshold crossing (hit) rate. The resulting likelihood ratio test is practical since the hit rate is both controllable and observable, unlike the false-alarm rate which is only controllable. The single-stage rate-constraint criterion is extended to the multi-stage case where a number of individual detectors are cascaded; each test is slower but more capable than the previous. A new parameter, the SLOC function (the slope of the Receiver Operating Characteristic Curve when plotted on log-log axes) is derived and seen to be critical to the performance of a detection system utilizing such a sequence of tests. Provided the SLOC numbers for the individual tests are properly ordered, the overall system detection performance under a wide range of criterion (Bayes, Neyman-Pearson, Maximum Mutual Information) is maximized when the rate-constraint criterion is used for all stages but the last. The resulting sequence of tests is again practical to control since each threshold depends only on local variables. TRUNCATED.
Keywords
Resource constrained detection;Rate constrained detection;Hypothesis testing
Report Number
URL-TR-93-01A;DREO-CR-93-657 — Contract Report (Final)
Date of publication
01 Mar 1993
Number of Pages
37
DSTKIM No
95-00832
CANDIS No
148608
Format(s):
Hardcopy;Document Image stored on Optical Disk

Permanent link

Document 1 of 1

Date modified: