Automatic detection and taxonomic identification of dolphin vocalisations using convolutional neural networks for passive acoustic monitoring

• DOI: 10.1016/j.ecoinf.2023.102291
• Publication Date: 2023-09-12T09:26:54Z
• Authors: Guilherme Frainer; Emmanuel Dufourq; Jack Fearey; Sasha Dines; Rachel Probert; Simon Elwen; Tess Gridley
• Keywords: Human echolocation; Delphinus delphis; Identification

A novel framework for acoustic detection and species identification is proposed to aid passive monitoring studies on the endangered Indian Ocean humpback dolphin (Sousa plumbea) in South African waters. Convolutional Neural Networks (CNNs) were used both of vocalisations tasks, performance was evaluated using custom pre-trained architectures (transfer learning). In total, 723 min data annotated presence whistles, burst pulses echolocation clicks produced by Delphinus delphis (~45.6%), Tursiops aduncus (~39%), Sousa plumbea (~14.4%), Orcinus orca (~1%). The best performing models detecting segments (spectral windows) two second lengths trained images with 70 90 dpi, respectively. model built a customised architecture achieved an accuracy 84.4% all test set, 89.5% high signal noise ratio. also correctly identified S. (96.9%), T. (100%), D. (78%) encounters testing dataset. developed designed based knowledge complex sounds it may assists finding suitable CNN hyper-parameters other or populations. Our study contributes towards development open-source tool assist long-term species, living highly diverse habitats, monitoring.