Application of wavelets and kernel methods to detection and extraction of behaviours of freshwater mussels

Some species of mussels are well-known bioindicators and may be used to create a Biological Early Warning System. Such systems use long-term observations of mussels activity for monitoring purposes. Yet, many of these systems are based on statistical methods and do not use all the potential that stays behind the data derived from the observations. In the paper we propose an algorithm based on wavelets and kernel methods to detect behaviour events in the collected data. We present our algorithm together with a discussion on the influence of various parameters on the received results. The study describes obtaining and pre-processing raw data and a feature extraction algorithm. Other papers which applied mathematical apparatus to Biological Early Warning Systems used much simpler methods and their effectiveness was questionable. We verify the results using a system with prepared tags for specified events. This leads us to a classification of these events and creating a Dreissena polymorpha behaviour dictionary and a Biological Early Warning System. Results from preliminary experiments show, that such a formulation of the problem, allows extracting relevant information from a given signal and yields an effective solution of the considered problem.