Classification of bird songs in water sites by using a unified method of neural network and Bayesian optimum discrimination

摘要

This article presents experimental results of classification of bird songs spectra in water sites, based on the previously proposed method that unifies the discriminative nonlinear transformation ability of neural networks and the Bayesian stochastically optimum discrimination. That is, when sample populations are overlapped to a low degree, classification is carried out only by the neural network, while they are to the high, the output vectors of the neural network are probabilistically classified by the Bayesian discrimination. The effectiveness and special features of the method are shown through classification experiments of 7 kinds of the bird songs.