Classification Method of Birdsong Based on Gabor_WT Feature Image and Convolutional Neural Network

In the classification of birdsong, the effect of traditional feature extraction methods and identifying birds through deep learning methods is still not ideal. Therefore, this paper proposes a birdsong classification method based on Gabor_Wt feature image and convolutional neural network. According to the two-dimensional wavelet spectrogram signal can present more detailed information characteristics of birdsong. The two-dimensional wavelet spectrogram is obtained by using wavelet transform from birdsong audio. The Gabor_Wt feature images are extracted texture information from wavelet spectrogram with 2D-Gabor filters. With the convolutional neural network structure, WT_CNN and Gabor Wt Cnn classification models are built. The experimental results show that the Gabor_Wt_Cnn method can achieve a classification accuracy of 94%, outperforming WT_CNN of 87% and random forest with MFCC features of 80%. The proposed method provides a new reference to birdsong classification.