The song of the humpback whale (Megaptera Novaeangliae) is one of the most varied vocal displays in the animal kingdom; it consists of complex, hierarchical structures and undergoes constant evolutionary change. Studying humpback whale song recorded using long-term acoustic recording devices requires highly skilled bioacousticians to listen to hours of audio recordings, identifying times in which humpback whales are vocalizing. This project aims to develop a machine learning algorithm able to automate the process of detecting the vocalizations based on pattern recognition using spectrogram images. Deep learning convolutional neural networks (CNNs) have made major advances in the previous decade, leading to a wide range of applications in image recognition. The detection of frequency modulated vocalizations by cetaceans is one of these applications. An existing algorithm to detect humpback whale song produced by Allen et al.  in a collaboration between the National Oceanic and Atmospheric Administration (NOAA) and Google (ResNET50 architecture), has achieved satisfactory results for data collected in the geographic area where it was trained (Hawaii), not however on data from the North Atlantic Ocean. This project features the retraining of the existing NOAA/Google detector on humpback whale song recordings from Scottish waters (and if successful the entire North Atlantic). Furthermore, different data augmentation techniques were used to artificially increase the variability within the training set with the goal of creating a detector, which is robust towards different noise environments.
 A. N. Allen, M. Harvey, L. Harrell, A. Jansen, K. P. Merkens, C. C. Wall, J. Cattiau, and E. M. Oleson. A Convolutional Neural Network for Automated Detection of Humpback Whale Song in a Diverse, Long-Term Passive Acoustic Dataset. Frontiers in Marine Science, 8, 2021.