With BuzzCam, we aim to develop a noninvasive system to monitor wild bee populations to gain further insight into their population density and behavior patterns. Currently, unscalable human observations are the standard for studying and identifying bee species. We propose using novel acoustic sensors and cameras to detect bumblebees and capture their buzzes, which are unique to each species. This multi-modal system could enhance conservation monitoring for bumblebees in their natural habitats.
We've designed the acoustic system from the ground up, paying special attention to reducing any noise artifacts that can influence the quality of the wildlife recordings. With material assistance from SAATI, we designed and tested a mechanical enclosure around our microphones to reduce wind noise while not significantly impacting the audio quality and system robustness to the elements. Our system also integrates a variety of environmental sensors and capabilities to run on-device machine learning algorithms to increase the duration of deployment in remote regions. All the data is efficiently stored on large-capacity SD cards provided by Kioxia but the systems have the ability to stream data wirelessly.
With generous support from the National Geographic Exploration Technology Lab, we deployed this system in March 2024 in Puerto Blest, Argentina, alongside National Geographic Explorers and professors Marina Arbetman and Cristian A. Villagra Gil. During that expedition, we collected over 100 hours of audio recordings and more than 10,000 annotated 10-second bee buzzes from both native and invasive bee species. In March 2025, we returned to the site to gather additional data and to field-test on-device machine learning models trained on the 2024 dataset.