Serial combination of multiple classifiers for automatic blue whale calls recognition

In this paper, we propose a serial combination architecture of classifiers for automatic blue whale calls recognition. Based on class’s best selection operator, the proposed system uses a best classifier for D call class followed by another one that efficiently discriminate the A and B calls. The first classifier uses the short-time Fourier transform to characterize the patterns, while the second uses the chirplet transform. Both classifiers are based on multi-layer perceptron neural network. The classification performance (95.55%) of the proposed system outperforms all tested single classifiers. The other advantages of the system are no requirement for adjusting a series of parameters and simple feature extraction.