Optimization of Deep Learning Models for Bird Identification in the Pantanal Using Embedded Devices

• Isis Milena Daron UFMT
• Julia Milioranza Gomes UFMT
• Thiago Meirelles Ventura UFMT / INPP

DOI: https://doi.org/10.5753/eniac.2026.30056

Resumo

Autonomous biodiversity monitoring in remote ecosystems such as the Pantanal requires on-device inference under strict resource constraints. To this end, deep learning models must demand reduced memory footprint and computational cost compared to standard architectures. This work evaluates post-training quantization, knowledge distillation, and pruning individually and in hybrid combinations applied to BirdNET v2.4 adapted for Pantanal bird species. The best-performing configuration achieved 96.72% accuracy at 3.63× compression, while hybrid pipelines reached 11.55× at 87.99%, with KD-first sequences outperforming pruning-first by up to 7.5 percentage points. The identified profiles support on-device bioacoustic inference without cloud dependency.

Referências

Abadi, M., Agarwal, A., Barham, P., Brevdo, E., Chen, Z., Citro, C., Corrado, G. S., Davis, A., Dean, J., Devin, M., Ghemawat, S., Goodfellow, I., Harp, A., Irving, G., Isard, M., Jia, Y., Jozefowicz, R., Kaiser, L., Kudlur, M., Levenberg, J., Mané, D., Monga, R., Moore, S., Murray, D., Olah, C., Schuster, M., Shlens, J., Steiner, B., Sutskever, I., Talwar, K., Tucker, P., Vanhoucke, V., Vasudevan, V., Viégas, F., Vinyals, O., Warden, P., Wattenberg, M., Wicke, M., Yu, Y., and Zheng, X. (2015). Tensor-Flow: Large-scale machine learning on heterogeneous systems. Software available from tensorflow.org.

Alho, C. J. R. and Sabino, J. (2011). A conservation agenda for the pantanal’s biodiversity. Brazilian Journal of Biology, 71:327–335.

Daron, I. M., Gomes, J., Ventura, T. M., Teles, L. K. S., and Oliveira, A. G. d. (2026). A systematic review of post-training optimization for embedded deep learning to support biodiversity monitoring in the pantanal. In Anais do Congresso de Inteligência Artificial de Mato Grosso, Cuiabá, MT, Brasil.

Gholami, A., Kim, S., Dong, Z., Yao, Z., Mahoney, M. W., and Keutzer, K. (2021). A survey of quantization methods for efficient neural network inference.

Gou, J., Yu, B., Maybank, S. J., and Tao, D. (2021). Knowledge distillation: A survey. International Journal of Computer Vision, 129(6):1789–1819.

Han, S., Mao, H., and Dally, W. J. (2015). Deep compression: Compressing deep neural networks with pruning, trained quantization and huffman coding.

Hinton, G., Vinyals, O., and Dean, J. (2015). Distilling the knowledge in a neural network.

Huang, Z., Tousnakhoff, A., Kozyr, P., Rehausen, R., Bießmann, F., Lachlan, R., Adjih, C., and Baccelli, E. (2024). Tinychirp: Bird song recognition using tinyml models on low-power wireless acoustic sensors.

Höchst, J., Bellafkir, H., Lampe, P., Vogelbacher, M., Mühling, M., Schneider, D., Lindner, K., Rösner, S., Schabo, D., Farwig, N., and Freisleben, B. (2022). Bird@Edge: Bird Species Recognition at the Edge, pages 69–86.

Jacob, B., Kligys, S., Chen, B., Zhu, M., Tang, M., Howard, A., Adam, H., and Kalenichenko, D. (2018). Quantization and training of neural networks for efficient integer-arithmetic-only inference. In 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition, page 2704–2713. IEEE.

Junk, W. J., da Cunha, C. N., Wantzen, K. M., Petermann, P., Strüssmann, C., Marques, M. I., and Adis, J. (2006). Biodiversity and its conservation in the pantanal of mato grosso, brazil. Aquatic Sciences, 68:278–309.

Kahl, S., Wood, C., and Klinck, H. (2025). Birdnet model v2.4.

Kahl, S., Wood, C. M., Eibl, M., and Klinck, H. (2021). Birdnet: A deep learning solution for avian diversity monitoring. Ecological Informatics, 61:101236.

Krishnamoorthi, R. (2018). Quantizing deep convolutional networks for efficient inference: A whitepaper. CoRR, abs/1806.08342.

LeCun, Y., Denker, J., and Solla, S. (1989). Optimal brain damage. In Touretzky, D., editor, Advances in Neural Information Processing Systems, volume 2. Morgan-Kaufmann.

Li, Y. and Ablavatski, A. (2021). Build fast, sparse on-device models with the new TF MOT pruning API. TensorFlow Blog. Accessed: 2025.

McFee, B., Raffel, C., Liang, D., Ellis, D., McVicar, M., Battenberg, E., and Nieto, O. (2015). librosa: Audio and music signal analysis in python. In Proceedings of the 14th Python in Science Conference, SciPy, page 18–24. SciPy.

Mohaimenuzzaman, M., Bergmeir, C., West, I., and Meyer, B. (2023). Environmental sound classification on the edge: A pipeline for deep acoustic networks on extremely resource-constrained devices. Pattern Recognition, 133:109025.

Silva, R. L. d., Jacinto, G., Véstias, M., and Duarte, R. P. (2026). Real-time bird audio detection with a cnn-rnn model on a soc-fpga. Electronics, 15(2):354.

Simonyan, K. and Zisserman, A. (2015). Very deep convolutional networks for large-scale image recognition.

Stowell, D. (2021). Computational bioacoustics with deep learning: a review and roadmap. CoRR, abs/2112.06725.

Wang, K., Yang, F., Chen, Z., Chen, Y., and Zhang, Y. (2023). A fine-grained bird classification method based on attention and decoupled knowledge distillation. Animals, 13(2):264.

Wood, C. M. and Kahl, S. (2024). Guidelines for appropriate use of birdnet scores and other detector outputs. Journal of Ornithology, 165(3):777–782.

Zhao, T., Xie, Y., Wang, Y., Cheng, J., Guo, X., Hu, B., and Chen, Y. (2022). A survey of deep learning on mobile devices: Applications, optimizations, challenges, and research opportunities. Proceedings of the IEEE, 110:334–354.