Abstract:

Given that the signals of crested ibis’s chirp are nonstationary, and the FFT cannot reflect transient signals or get rid of the basis function, we propose a feature extraction algorithm based on the HillberHuang transform (HHT) MFCC parameters. Each frame of the intrinsic mode function is obtained through the EMD decomposition of the signals, on which the HT transform is then conducted. The Hilbert marginal spectrum of the frequency synthesis passes though the Mel filter, and then the logarithmic energy is obtained. The MFCC coefficients are improved after the DCT transform, and the crested ibis individual can be recognized through the Gauss mixture model in pure and noisy environment respectively under different signaltonoise ratios. Experimental results show that the improved algorithm can reflect transient chirp signals better, and the average recognition rate of the crested ibis individuals is also increased by 4%.

Key words: crested ibis individual identification;FFT;MFCC algorithm;HHT algorithm