Senior Theses 2008
Daniel Widener (2008) Comparison of three algorithms for automatic detection of bottlenose dolphin Tursiops truncatus whistles
Faculty Advisor: Shannon Gowans
Automatic detection algorithms are invaluable tools for dealing with the large quantities of data generated in passive acoustic monitoring studies. Multiple approaches to automatic detection are possible, but I focused on designing algorithms to detect bottlenose dolphin (Tursiops truncatus) whistles based on frequency and slope characteristics. After examining an initial automatic detection algorithm, I wrote two more algorithms designed to improve detection of bottlenose dolphin whistles in low signal-to-noise ratio data. Using more comprehensive methods of searching in the second and third algorithms improved detection rates greatly from the first algorithm, as detection rates were 8%, 19%, and 40% in the first, second, and third algorithms respectively. Though detection rates increased significantly, false detections remained a problem, as false detection rates were always several times higher than actual detection rates. In particular, all three algorithms are easily confused by noise such as boat noise and fish chorus noise. Unless further modifications are able to reduce false detection rates in the third algorithm, it is unlikely that this approach will be feasible for detection of T. truncatus whistles in low signal-to-noise ratio data.