Lightweight Models for Edge-based Passive Acoustic Monitoring of Pollinators

ulaval.dat were derived from the fourth dataset listed under Related works. Galen, C., Susič, D., Gradišek, A., Gil, C. V., Schul, J., Robinson, V. C., ... & Arbetman, M. P. (2026). All the flight we cannot see: Using passive acoustic monitoring to track the world's largest bumble bee, Bombus dahlbomii. Ecological Entomology, 51(3), 442-453. patagonia.dat were derived from the second and third dataset listed under Related works. Chwalek, P., Kuronaga, M., Zhu, I., Montague, S., Campopiano Robinson, V., Lohrmann, J., ... & Arbetman, M. (2025). High-Res Acoustic and Environmental Data to Monitor Bombus dahlbomii Amid Invasive Species, Habitat Loss. Scientific Data, 12(1), 548. ohio.dat were derived from the first dataset listed under Related works. Forrester, K. C., Lin, C. H., & Johnson, R. M. (2024). Measuring factors affecting honey bee (Hymenoptera: Apidae) attraction to soybeans using bioacoustics monitoring. Journal of Insect Science, 24(2), 20. jsi.dat were derived from the fifth dataset listed under Related works. Susič, D., Robinson, J. A., Bevk, D., & Gradišek, A. (2025). Acoustic monitoring of solitary bee activity at nesting boxes. Ecological Solutions and Evidence, 6(3), e70080. Trained models are released under the MIT License. MobileNetV2 and ShuffleNetV2 models were fine-tuned from pretrained weights provided by torchvision. ResNet9 and BeeNet were trained from scratch.