Trainable Linear Filters for the Automated Detection of Mines in Side Scan Sonar Images

PDF

Authors
  1. Kessel, R.T.
Corporate Authors
Defence Research Establishment Atlantic, Dartmouth NS (CAN)
Abstract
The multibeam side scan sonar images used for sea mine hunting look rather like aerial photographs of the seafloor, but there are important differences. The sonar's automatic gain control, the consistent highlight-shadow signature progression of objects, and the absence of vanishing points, considerably reduce the degrees of freedom in an object's appearance, making sonar images better suited for linear detection schemes than optical photographs generally are. This paper demonstrates a trainable linear filber for the detection of mine-like objects in side scan sonar images. The training is fast and flexible, using actual or synthetic sonar images to prescribe what image patterns are to be detected and rejected, and with a sliding filter window of any shape to focus more particularly on important features. Measures of performance such as the probabilities of detection and false alarm, and the limits of trained discrimination and window size are derived. The method is demonstrated for mine-like targets in actual sonar images.
Keywords
Highlight/shadow;Target identification;Classification algorithms;Target shapes;Automatic target detection
Report Number
DREA-TM-1999-171 — Technical Memorandum
Date of publication
01 Dec 1999
Number of Pages
50
DSTKIM No
CA001640
CANDIS No
513528
Format(s):
Hardcopy;Document Image stored on Optical Disk

Permanent link

Document 1 of 1

Date modified: