Home // IARIA Congress 2024, The 2024 IARIA Annual Congress on Frontiers in Science, Technology, Services, and Applications // View article
Authors:
Maria Jose Guerrero Muriel
Santiago Taborda Diosa
Juan Manuel Daza Rojas
Claudia Victoria Isaza Narvaez
Keywords: Machine learning, Deep learning, Clustering, Bioacoustics, Soundscape, Species identification.
Abstract:
Passive Acoustic Monitoring (PAM) using computational intelligence techniques offers new avenues for biodiversity conservation, particularly in identifying and monitoring species within tropical ecosystems. While various methods exist for animal sound identification, a comprehensive understanding of their advantages and disadvantages is often lacking. This work evaluates five methods for automatically identifying species vocalizations across different taxonomic groups using an acoustic dataset from a Colombian agricultural ecosystem. We conducted a comparative analysis of supervised techniques, including Convolutional Neural Networks (CNN), Random Forest (RF), and Support Vector Machine (SVM), as well as unsupervised methods such as spectral clustering, DBSCAN, and the Learning Algorithm for Multivariate Data Analysis (LAMDA) 3pi, evaluating their species detection performance through the F1-Score metric. Our research underscores the critical role of methodological selection in achieving accurate species identification. Furthermore, this study advances the understanding of clustering interpretation, illustrating its potential beyond bioacoustic studies. It presents how unsupervised learning techniques can be valuable in scenarios characterized by limited labeled data, common in tropical ecosystems, and high uncertainty regarding the number of clusters obtained. This approach facilitates the exploration of prototype patterns, aiding species association and potentially extending to other areas requiring insight into unidentified clusters. This study offers valuable insights into selecting suitable tools for bioacoustic studies, emphasizing the need for comprehensive input preparation for model training. The findings underscore the potential of PAM and computational strategies in furthering biodiversity research and conservation efforts, effectively addressing the challenges of species identification and clustering interpretation.
Pages: 126 to 133
Copyright: Copyright (c) IARIA, 2024
Publication date: June 30, 2024
Published in: conference
ISBN: 978-1-68558-180-0
Location: Porto, Portugal
Dates: from June 30, 2024 to July 4, 2024