A Study On Technology Of Individual Recognition Based On Stridulation Of Crested Ibis

Crested ibis belongs to the first-class protected animal in our country, and it is rare bird. Since its stridulation contains rich biological information, analyzing and studying the stridulation of crested ibis is of great practical significance for understanding its habits and further protection.Based on the fundamental technology of sound recognition, this thesis analyzes pre-process, feature extraction and pattern matching of the stridulation of crested ibis.For feature extraction, based on the problems that extraction of the high frequency is easy to lose after the traditional Mel Frequency Cepstrum Coefficient(MFCC) extracts the sound, which makes it hard to represent the signal of the stridulation of crested ibis,improving method is put forward. This method adopts Wavelet packet transformation and high frequency weighted to subtly divide the signal of the stridulation of crested ibis so that spectrum energy of the high frequency of the stridulation can be strengthened. Compared with traditional MFCC, this method can obtain feature parameter of better consequence. In terms of recognition, for the problems that neurons are not fully taken advantage of and are sensitive to initial values existing in Learning Vector Quantization(LVQ) neural network, this thesis applies genetic algorithms to optimizing the initial values. And meanwhile, LVQ is used to finish studying the given accuracy to rapidly obtain the optimal weight vector, which makes the convergent speed of the recognition system quicker, and avoids local minimum value, and the recognition rate is improved.The recognition system of crested ibis based on LVQ is realized, and a comparative experiment between algorithms in this thesis and common recognition algorithms is done in a small file base containing stridulation of ten crested ibis. The resultsdemonstrate that the improved LVQ algorithms further improves the recognition of the system, obtains better recognition effect, and provides a quick and accurate method for future study.