Support vector machines for classification of underwater targets in sidescan sonar imagery


  1. Couillard, M.
  2. Fawcett, J.A.
  3. Myers, V.L.
  4. Davison, M.
Corporate Authors
Defence R&D Canada - Atlantic, Dartmouth NS (CAN);Defence R&D Canada - Centre for Operational Research and Analysis, Ottawa ON (CAN)
High-frequency sidescan sonars have become a fundamental tool for modern mine hunting operations. They produce quality images of the seabed and a high probability of detection can be achieved. A crucial classification phase is then needed to accurately identify these contacts as harmless or as potential underwater mines. To ensure the security of the follow-on traffic, all mines have to be identified accurately. At the same time, to avoid delays in the minefield clearing operations, clutter should not be misclassified as mines. For this contact identification task, multiple classification tools are available. This technical memorandum focuses on a powerful classification tool: support vector machines. A comprehensive introduction to support vector machines is provided and their usefulness for the classification of underwater objects in sidescan sonar imagery is investigated. The database used in this study is made of real sidescan sonar images collected during the CITADEL sea trial held at the NATO Undersea Research Center in October 2005. It is shown that this classification tool yields excellent classification performances when shadow-based features are used. These performances increase significantly when highlight-based features are added. It is also shown that the Ridge regression approximation is faster than quadratic optimization for large dataset and yields a comparable performance.

Il y a un résumé en français ici.

Report Number
DRDC-ATLANTIC-TM-2008-190 — Technical Memorandum
Date of publication
01 Nov 2008
Number of Pages
Electronic Document(PDF)

Permanent link

Document 1 of 1

Date modified: