Regional Adaptation of Deep Learning Models for Bird Species Classification in the Pantanal

• Julia Milioranza Gomes UFMT
• Isis Milena Daron UFMT
• Thiago Meirelles Ventura UFMT / INPP
• Allan Gonçalves de Oliveira UFMT
• Marinêz Isaac Marques UFMT / INPP / INCT-INAU / CO.BRA
• Karl-Ludwig Schuchmann UFMT / INCT-INAU / CO.BRA / Zoological Research Museum A. Koenig

DOI: https://doi.org/10.5753/eniac.2026.31242

Resumo

Despite being one of the world’s most biodiverse wetlands, the Pantanal remains underrepresented in global bioacoustic datasets. This study develops a Pantanal-specific bird classification model using transfer learning from BirdNET and Xeno-Canto recordings. An evaluation of acoustic representations (MFCCs, Mel-spectrograms, and linear spectrograms) revealed the limitations of training CNNs from scratch on restricted regional data, establishing the necessity for transfer learning. Consequently, the transfer learning model circumvented these representation bottlenecks and outperformed the original BirdNET, reaching a weighted F1-score of 0.95 on the Pantanal validation set and 0.92 on an external dataset, indicating limited geographic overfitting. Results highlight the effectiveness of regional fine-tuning for automated bird monitoring in tropical environments.

Referências

Clarfeld, L. A., Gieder, K. D., Abrams, R., Bernier, C., Cahill, J., Staats, S., Wixsom, S., and Donovan, T. M. (2025). Two-stage models improve machine learning classifiers in wildlife research: A case study in identifying false positive detections of ruffed grouse. Ecological Informatics, 89.

Dematties, D., Rajani, S., Sankaran, R., Shahkarami, S., Raut, B., Collis, S., Beckman, P., and Ferrier, N. (2024). Acoustic fingerprints in nature: A self-supervised learning approach for ecosystem activity monitoring. Ecological Informatics, 83.

Duarte, A., Weldy, M. J., Lesmeister, D. B., Ruff, Z. J., Jenkins, J. M., Valente, J. J., and Betts, M. G. (2024). Passive acoustic monitoring and convolutional neural networks facilitate high-resolution and broadscale monitoring of a threatened species. Ecological Indicators, 162.

Gregory, R. D. and van Strien, A. (2010). Wild Bird Indicators: Using Composite Population Trends of Birds as Measures of Environmental Health. Ornithological Science, 9(1):3 – 22.

Heinrich, R., Rauch, L., Sick, B., and Scholz, C. (2025). Audioprotopnet: An interpretable deep learning model for bird sound classification. Ecological Informatics, 87.

Hooper, D. U. e. a. (2005). Effects of biodiversity on ecosystem functioning: a consensus of current knowledge. Ecological Monographs, 75:3–35.

Kahl, S., Wood, C., and Klinck, H. (2025). Birdnet model v2.4 (version 1) [figure].

Kahl, S. e. a. (2021). Birdnet: A deep learning solution for avian diversity monitoring. Ecological Informatics, 61.

Kazeneza, M., Bosman, A. S., Amenyedzi, D. K., Hanyurwimfura, D., Ndashimye, E., and Vodacek, A. (2025). Balancing complexity and performance of machine learning models for avian pests sound detection in agricultural environments. 13.

Michaud, F., Sueur, J., Cesne, M. L., and Haupert, S. (2023). Unsupervised classification to improve the quality of a bird song recording dataset. Ecological Informatics, 74.

Park, D. S., Chan, W., Zhang, Y., Chiu, C.-C., Zoph, B., Cubuk, E. D., and Le, Q. V. (2019). SpecAugment: A Simple Data Augmentation Method for Automatic Speech Recognition. In Interspeech 2019, pages 2613–2617.

Purwins, H., Li, B., Virtanen, T., Schlüter, J., Chang, S.-Y., and Sainath, T. (2019). Deep learning for audio signal processing. IEEE Journal of Selected Topics in Signal Processing, 13(2):206–219.

Pérez-Granados, C. (2023). Birdnet: applications, performance, pitfalls and future opportunities. Ibis, 165:1068–1075.

Stowell, D. (2022). Computational bioacoustics with deep learning: a review and roadmap. PeerJ, 10:e13152.

UNESCO World Heritage Centre (2000). Pantanal Conservation Area.

Ventura, T. M., Ganchev, T. D., Pérez-Granados, C., de Oliveira, A. G., de S. G. Pedroso, G., Marques, M. I., and Schuchmann, K. L. (2024). The importance of acoustic background modelling in cnn-based detection of the neotropical white-lored spinetail (aves, passeriformes, furnaridae). Bioacoustics, 33:103–121.

Wang, Y., Chen, A., Li, H., Zhou, G., Yi, J., and Zhang, Z. (2023). A hierarchical birdsong feature extraction architecture combining static and dynamic modeling. Ecological Indicators, 150.

Xie, J. and Zhu, M. (2022). Sliding-window based scale-frequency map for bird sound classification using 2d- and 3d-cnn. Expert Systems with Applications, 207.

Xie, S., Lu, J., Liu, J., Zhang, Y., Lv, D., Chen, X., and Zhao, Y. (2022). Multi-view features fusion for birdsong classification. Ecological Informatics, 72.

Zhang, C., Li, Q., Zhan, H., Li, Y. F., and Gao, X. (2023). One-step progressive representation transfer learning for bird sound classification. Applied Acoustics, 212.