A Novel Dolphin Whistle Classifier Based on Attention-DenseNet

Traditional whistle classification algorithms need complex preprocessing operations to classify dolphin whistle signals. Moreover, conventional methods cannot fully extract features. To solve the above problems, this paper proposes an Attention-DenseNet for accurate dolphin whistle classification. This model fuses features to learn whistle characteristics from multiple perspectives and introduces the self-attention module and the central loss function to assist the network in improving the classification performance. The experimental results show that the proposed model can fully exploit the whistle signal features to avoid information loss and improve classification accuracy.