Using cubic splines interpolation to enhance classification of marine mammals signal

Abstract

This paper presents a model for processing marine mammal (MM) signals using Wavelet transform combined with cubic splines interpolation before classifying the signals by Convolution Neural Network (CNN). The representation when interpolating the data can limit information loss after the raw data has been transformed, highlight the relationships between the characteristic frequencies of each species, achieving over 10% increase in classification accuracy on real datasets compared with the same CNN. The proposed representation has shown to be effective in classifying marine mammal signals which are sensitive to the traditional Fast Fourier transform (FFT) and Short time Fourier transform (STFT) solutions.