Research library

buzzdetect: an open-source deep learning tool for automated bioacoustic pollinator monitoring

Long-term pollinator studies are limited by the cost and poor scalability of traditional observations such as sweep netting and pan trapping. This work introduces buzzdetect, an open-source tool that applies deep learning to passive audio recordings to distinguish insect flight buzzes from environmental noise. At a precision of 95%, the accompanying model achieved 27% sensitivity. Applied to recordings from pumpkin, watermelon, mustard, and soybean fields, the system revealed differences in the timing and intensity of foraging activity. The authors provide the source code and current models on GitHub and archive the data and code required to reproduce the manuscript results on Zenodo.

Download PDF

Publication details

Authors
Luke E. Hearon, Lillian H. P. Johnson, James Underwood, Chia-Hua Lin, Reed M. Johnson
Organizations
🇺🇸 The Ohio State University🇺🇸 Dartmouth College
Year
2025
Type
Preprint

Relevancy to Gratheon

buzzdetect contributes a reproducible software and data pipeline for processing long-duration field audio, a capability that can complement Gratheon's in-hive acoustics with monitoring of foraging activity around an apiary. Its public code, models, and Zenodo archive make it useful for evaluating audio segmentation, deployment workflows, and transfer learning on Gratheon recordings. The reported operating point must be interpreted carefully: 95% precision minimizes false buzz detections, but 27% sensitivity misses most buzz events. It is therefore a practical open baseline and data resource, not yet a drop-in production counter.

Code and data

Source and access

The preprint is available from bioRxiv under the Creative Commons Attribution 4.0 license.