59. DETECTION AND CLASSIFICATION OF MINE LIKE OBJECTS IN SIDE SCAN SONAR IMAGERY

Department: Computer Science & Engineering
Faculty Advisor(s): Ryan Kastner | Serge Belongie

Primary Student
Name: Christopher M Barngrover
Email: cbarngro@ucsd.edu
Phone: 858-822-3634
Grad Year: 2013

Abstract
The task of detecting mine like objects (MLO's) in side scan sonar (SSS) imagery has a profound impact on our military operations. The current process relies on subject matter experts to spend time analyzing sonar images in search of any MLO's. An automated approach would carry substantial time and cost benefits. The automation concept has been heavily researched through a signal processing point of view. This poster shows the techniques that have dominated the research in the past and posits the use of common computer vision algorithms to address this problem. With this new approach in mind, we consider known feature types and the boosting approach to machine learning on our synthetic database of side scan sonar images. The synthetic database is produced from real world sonar images of various mine types. The results of these algorithms reinforce the difficult nature of this problem and show the potential for utilizing the features and methods prominent in computer vision approaches to optical imagery.

« Back to Posters or Search Results