Listening to the Hive

In Improved Monitoring of Honey Bee Colony Strength via Audio IoT Sensors, Modulation Tensorgrams and Recurrent Neural Networks, Mahsa Abdollahi and her colleagues explore using microphones and machine learning to estimate the strength of a bee colony without opening the hive.

“Accurate, generalizable, and robust acoustic monitoring of honey bee colony strength is possible.”

As a beekeeper, or perhaps more accurately a bee shepherd, I have often thought IoT could be useful for monitoring hive health and keeping track of what is happening while I am away. I just have not had enough time to explore it properly.

This study used more than 3,000 hours of audio collected from nine Langstroth hives over two years. Microphones inside the brood boxes recorded 15-minute samples every 30 minutes, while manual inspections measured colony strength by counting frames substantially covered with bees.

Instead of listening only for how a hive sounds at one moment, the researchers analyzed how its frequencies and rhythms changed over time. Their best model reduced the average error and generalized better when tested against hives it had never encountered before.

The system is also small and fast enough to be interesting outside a laboratory. The best-performing model was only 4.5 MB and completed inference in under five milliseconds on an M1 Mac, which the researchers say makes on-hive edge deployment plausible.

There are still limitations. The dataset came from only nine hives at one urban apiary, and weather, season, hive configuration, and time of day may influence the sounds being measured. More diverse testing will be needed before this can replace experienced eyes inside a hive.

Still, this is really awesome. A small device combining audio with temperature, humidity, weight, and perhaps a camera at the entrance could provide a remarkably detailed picture of a colony without constantly disturbing it.

The bees may already be telling us how the hive is doing. We are only beginning to learn how to listen.

Read the full paper, Improved Monitoring of Honey Bee Colony Strength via Audio IoT Sensors, Modulation Tensorgrams and Recurrent Neural Networks.