The by-catch of small cetaceans in fishing nets has been identified as a widespread problem, but attempts to reduce this require an understanding of the way these animals behave around the nets. One of the problems with assessing changes in behaviour between encounters is the difficulty of identifying individuals. Acoustic identification techniques overcome some of the problems associated with visual ID, and field research has shown that the presence of a sonobuoy and hydrophone have no effect on dolphin behaviour in the field. Dolphins produce whistles that can be used for identification, although current theory suggests these identify small groups rather than individuals. Novel algorithms have been developed to detect and process these tonal whistles, and their characteristic time-frequency-intensity contours extracted from the raw signals. Feature extraction techniques were developed for the contours based on timefrequency 'shape' of the contours, allowing a syntactic pattern recognition approach based around hidden Markov modelling to be employed for classification. The algorithms have enabled the whistles from concurrent whistles to be separated and analysed. Contours of 101 wild bottlenose dolphin whistles were successfully characterised. Analysis of the resulting classes indicated one group occurring only once and two other groups occurred twice but on different days. Another study was conducted of three groups of common dolphin, with a total of 49 recorded whistles analysed. The first group was found to contain whistles significantly different to either of the other two, although neither similarity nor dissimilarity could be inferred on the second and third. Further analysis suggested there were indeed two separate groups of dolphins for the last two groups, but that there was a period of overlap in their recording. A significant difference could be found between them once certain classes were re-assigned. It should be possible to apply these same techniques to a wider range of odontocete species, since most of those studied have been found to exhibit similar whistles. The tasks of whistle detection, isolation, and encoding can be applied automatically by computer with no loss of identity information, and these encoded contours can subsequently be quantitatively classified by their shape.
| Identifer | oai:union.ndltd.org:bl.uk/oai:ethos.bl.uk:242932 |
| Date | January 1997 |
| Creators | Sturtivant, Christopher R. |
| Publisher | Loughborough University |
| Source Sets | Ethos UK |
| Detected Language | English |
| Type | Electronic Thesis or Dissertation |
| Source | https://dspace.lboro.ac.uk/2134/6761 |
Page generated in 0.0021 seconds