EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

Classifying Wildlife Acoustic Signals using a Deep Learning Approach

Sunila Sheikh, Umer Rashid

International Journal of Innovations in Science and Technology · 08.05.2026 · 324

DOI: 10.33411/ijist/1781 ↗

Kurz gefasst

Fachartikel in der Zeitschrift International Journal of Innovations in Science and Technology, online erschienen am 08.05.2026. Verfasst von Sunila Sheikh und 1 weiteren Autor:innen. Die Arbeit nennt 35 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Monitoring wildlife and environmental conditions in national parks is essential for ecological research, biodiversity conservation, and public safety. This study proposes a contextual sound-based monitoring framework that addresses the limitations of vision-based systems in low-light and occluded environments commonly found in wildlife areas. The proposed approach integrates a hybrid deep learning architecture combining Convolutional Neural Networks (CNNs) for spatial feature extraction and a Bidirectional Gated Recurrent Unit (BiGRU) for temporal sequence modeling, along with a fuzzy logic decision layer for high-level contextual interpretation. To ensure diversity and robustness, multiple open-source datasets, including ESC-50, UrbanSound8K, FSC22, and Scream/Non-Scream datasets, are preprocessed, harmonized, and merged into a unified dataset comprising 15,811 audio clips across 16 low-level sound classes. The dataset includes alarming sounds, representing complex acoustic environments relevant to wildlife and park monitoring. The model employs a hierarchical classification strategy. Firstly, the CNN-BiGRU network performs low-level sound event classification, and then a fuzzy inference system maps the outputs into four high-level contextual categories: Illegal Activity, Human Distress, Natural Hazard, and Safe Activity. Experimental results demonstrate strong performance, achieving an accuracy of 95.80%, precision of 95.95%, recall of 96.14%, weighted F1-score of 95.80%, and ROC-AUC of 99.67% on the UrbanSound8K dataset. With an accuracy of 91.30%, precision of 88.11%, recall of 86.23%, weighted F1-score of 87.91%, and ROC-AUC of 99.50%, the model maintains competitive performance on the harmonized dataset with a difference of less than 5% across evaluation metrics. These findings highlight the effectiveness of contextual sound analysis in enhancing situational awareness and supporting intelligent surveillance systems for wildlife and environmental monitoring.

Zitieren

APA

Sheikh, S., & Rashid, U. (2026). Classifying Wildlife Acoustic Signals using a Deep Learning Approach. International Journal of Innovations in Science and Technology, 324. https://doi.org/10.33411/ijist/1781

Vancouver

Sheikh S, Rashid U. Classifying Wildlife Acoustic Signals using a Deep Learning Approach. International Journal of Innovations in Science and Technology. 2026:324. doi:10.33411/ijist/1781

BibTeX
@article{Sheikh2026,
  author = {Sheikh, Sunila and Rashid, Umer},
  title = {Classifying Wildlife Acoustic Signals using a Deep Learning Approach},
  journal = {International Journal of Innovations in Science and Technology},
  year = {2026},
  pages = {324},
  doi = {10.33411/ijist/1781}
}
Kostenlos starten →