Biomathematics and Statistics Scotland, Rowett Research Institute, Greenburn Rd., Aberdeen, AB21 9SB, Scotland, UK firstname.lastname@example.org
Wavelet techniques have been around since the start of the 1980s, but it has only been in the last 10 years that they have really come into their own, and have been applied in many different fields. They were developed to rival the well known Fourier analysis which tells us the dominant frequencies present in a signal. The advantage of wavelets is that they are able to tell what the dominant frequencies are and when they occur, thus providing more information than Fourier.
It is therefore believed that with more information, wavelets are better able to characterise and distinguish between signals that Fourier analysis techniques would not be able to tell apart. Previously, much of my research has been in image analysis and the automatic recognition of crop varieties from digital photographs. This was a particularly difficult problem due to the huge amount of variation in images from the same variety. Images, made up of thousands of individual pixels, contain large amounts of information, and we used the wavelet transform to capture the most important sources of textural variation and the structure in the pixel intensities.
We were introduced to haddock sounds by biologists at the Fisheries Research Services Marine Laboratory in Aberdeen. They were recording haddock knocks to understand the relationship between sound production and mating behaviour. Due to the volume of recorded sounds, they required some method of automatically identifying the individual haddock from the sounds produced. Knowing which fish produced a given sound would then allow the biologist to closely monitor the fishs behaviour.
Wavelet analysis of the haddock knocks has proved very successful, and including further information, such as the fact that haddock tend to emit long series of knocks, increases the recognition rates. Similar techniques have been used to discriminate between cod, pollack and haddock.
Ideally it would be nice to develop these techniques to identify different haddock and different species in real time. This would be particularly useful for scientists or fishermen at sea, where such a tool would simplify the search for spawning grounds of a particular fish species. The methodology is not restricted to classifying fish sounds, and could be applied to discriminate between any kind of source which produces sound or a similar wave-like signal.