The Raspberry Pi-Powered TensorFlow "Honeybee Booth" Keeps an Electronic Eye Out for Murder Hornets

Beekeeper and developer Sean Cusack has published software that uses a machine learning system to identify visiting insects and warn when a hive has caught the attention of the giant Asian hornet — also known as the "murder hornet" — with up to 98 percent accuracy.

"Last summer I used a Raspberry Pi, motion sensor, and camera to takes pictures of anything entering/leaving my honeybee hives. I called it 'The Honeybee Booth,'" Cusack writes. "The goal of the project was to use some machine learning AI to detect and count varroa mites and alert me with the results in real time. I figured if my device is simple and duplicable enough, then the community can easily replicate my device, install my code, and plug into this global monitoring dashboard."

"Given the situation with the Giant Asian Hornet, I've been working quickly to adapt this device to detect them as well. Things are a bit rough around the edges but my device is fully functional and capable of detecting the Giant Asian Hornet with about a 96% to 98% accuracy. More model training will be needed in the coming months."

The core of the platform is a Raspberry Pi of any model, with the Raspberry Pi Zero W consuming the least power but taking longer to process imagery, with an Arducam five megapixel camera, a VCNL4010 motion sensor connected to the Raspberry Pi's GPIO header, and a white LED to serve as a flash during image capture. The hardware is housed in a 3D printed enclosure.

On top of that is an object recognition model built using Python and TensorFlow, with all classification occurring locally on the Raspberry Pi itself. Telemetry data, meanwhile, is shared via Azure IoT Central.

The full source code, written in partnership with fellow developer Reuben Cleetus, and 3D model files are available on Cusack's GitHub repository under an unspecified license.