A probabilistic theory for the design of optimal linear discriminators for the automated detection of objects in sidescan sonar images


  1. Kessel, R.T.
Corporate Authors
Defence Research Establishment Atlantic, Dartmouth NS (CAN)
Computerized pattern recognition can be used to help a sonar operator locate underwater objects in sidescan sonar images. The theory behind several linear discriminators is presented here with a view to improving this automated detection. The discriminators are optimal insofar as they maximize the detection performance as defined under the Neyman-Pearson design criteria, with adjustments made to those criteria to suit the prior knowledge of both the objects sought and the local seafloor clutter. The emphasis throughout is on sea mine detection in naval operations. The theory gives practical insight and direction for the mine detection problem, showing, for instance, 1) what kind of data should be extracted from target and clutter image libraries to get optimal detection performance; 2) that the matched filter, favored for its implicity, is just one of several optimal linear discriminators resulting in this case when nothing is assumed about the local seafloor clutter; and 3) that prior de-meaning of images will in some cases improve detection performance.
Classification algorithms;Target identification;Target shapes;Automatic target detection;Clutter reduction;Computer aided classification;Computer aided detection;Optimal discriminator
Report Number
DREA-TM-2000-021 — Technical Memorandum
Date of publication
01 Mar 2000
Number of Pages
Hardcopy;Document Image stored on Optical Disk

Permanent link

Document 1 of 1

Date modified: