Applying Machine Learning Method To Identify Indo-Pacific Humpback Dolphin Click Signals

Accurate and efficient identification of echolocation click signals of cetaceans plays important role in conservation studies. However, it is challenging to analyze large amounts of acoustic data by traditional manual analysis methods. In this study, two supervised machine learning algorithms (the Alexnet neural network and Libsvm) were trained to automatically identify echolocation clicks of Indo-Pacific humpback dolphins. Wavelet transform was implemented to reflect the characteristics of click signals of Indo-Pacific humpback dolphins in time-frequency images, and these images were fed into the network for training. The better performance was reported by the Alexnet neural network, the click identification accuracy of which is up to 99.7%. This study shows that the Alexnet neural network method is more efficient and available in Indo-Pacific humpback dolphin clicks identification. When this method is mature enough in the near future, then it can be applied in AUV to identify Indo-Pacific humpack dolphins on line.