Publication
Speeding Up Bioacoustic Data Analysis: Fine-Tuning Deep Models with Active Learning for Efficient Wildlife Detection
Hannes Kath; Thiago Gouvea; Daniel Sonntag
In: GoodIT '25: Proceedings of the 2025 International Conference on Information Technology for Social Good. ACM International Conference on Information Technology for Social Good (GoodIT-2025), located at GoodIT-2025, September 3-5, Antwerpen, Belgium, Association for Computing Machinery, 9/2025.
Abstract
Biodiversity loss is accelerating, and evidence-based ecosystems management requires up-to-date and reliable quantitative data on habitat biodiversity.
Passive acoustic monitoring (PAM) has emerged as a key technology for scalable wildlife monitoring.
While PAM enables large-scale data acquisition, analyzing the vast amount of recorded data remains a challenge.
This work explores the use of transfer learning and active learning for efficient PAM data analysis by investigating the impact of fine-tuning a state-of-the-art transfer learning model and applying active learning.
Our results show that the highest performance-to-computation-time ratio is achieved when using fine-tuning and active learning with a dynamically increasing batch size for sample selection.
This work lays the groundwork for future research focused on developing an efficient, user-friendly PAM analysis tool for scalable data analysis, ultimately promoting the use of PAM in wildlife monitoring.