Advancing Dolphin Acoustic Monitoring: A Comprehensive Whistle Classification Framework

Dolphins are widely recognized as intelligent marine mammals with sophisticated communication and echolocation. Accurately classifying their whistles is essential for understanding how they communicate and for tracking population size, structure, and distribution. Here, we assemble a large, high-quality dataset of dolphin whistle signals collected at the Chimelong Ocean Kingdom, including a whistle type not previously available to researchers. We then explore Convolutional Neural Networks (CNNs) for classifying whistles of the Indo-Pacific bottlenose dolphin (Tursiops aduncus), testing 5 CNN architectures to analyse the signals. Model performance is reported using mean Average Precision (mAP), showing that CNN approaches can reliably separate different whistle classes. To probe robustness, we also introduce noise at defined SNR levels to increase testing complexity and assess the stability of the classifier. We use Bellhop for channel simulation to construct the channel impulse response. The simulated data can be used as augmented data to add to the original data training set. The results did indicate that this can enhance the robustness of the classification model. This work provides valuable tools for marine biologists and researchers specialising in animal acoustics, enhancing the understanding of dolphin communication. It also contributes to the conservation and management efforts of dolphin populations, offering significant insights into their behaviour and ecological needs.