Development Progress
A running log of what's working, what's being built, and what's coming next.
Current build: Prototype v0.3
Assembled and running in test mode. This version includes:
Thermal imaging
Captures brood nest temperature frames every 10 minutes. Working reliably in controlled tests. Need to validate performance in humid hive conditions over weeks, not days.
Humidity and temperature sensing
Integrated and logging data. Sensors are accurate, but placement inside the hive matters more than I expected—small position changes affect readings significantly.
Acoustic monitoring
Audio capture is functional. Lots of noise to filter out (bee wing beats, environmental sounds, wind interference). Working on signal processing to isolate meaningful patterns.
Video capture at hive entrance
Early pipeline built. Captures frames at intervals. Next step is building bee-counting models and traffic pattern detection, currently manual analysis only.
Web dashboard
Live data display, basic charting, historical trends. It works, but needs better data visualisation and alert configuration. Dashboard is usable, not polished.
Over-the-air updates
Firmware update system tested internally. This means I can push updates remotely once units are deployed in hives. Critical for fixing bugs without physically retrieving devices.
What I'm working on now
Machine learning pipeline: Gathering training data for thermal anomaly detection. The plan is to train models that can flag unusual temperature patterns automatically (e.g. brood nest cooling overnight, abnormal heat distribution). Still early—need more diverse data from different hive conditions.
Battery optimisation: Current prototype targets 2-3 months between charges, but that depends heavily on camera usage. Testing different power profiles to balance data collection with battery life. Solar recharging is a possibility for future iterations.
Acoustic signal processing: Filtering out noise and isolating meaningful hive sounds. Experienced beekeepers can hear when a hive is queenless or stressed, trying to automate that same detection. It's harder than it sounds.
Field durability: Current enclosure works in controlled environments. Need to test whether it holds up in real hives: propolis buildup, condensation, temperature extremes, bees chewing things they shouldn't. January field trials will reveal a lot.
Next milestones
January 2026: Field test with three hives
Deploy prototype units in real hives for extended monitoring. This is the first proper test of whether the hardware, sensors, and power system actually work outside a lab. Expect things to break and need iteration.
February–March 2026: Data collection and model training
Gather thermal, acoustic, and environmental data from field-deployed units. Start training ML models for anomaly detection. Validate accuracy against manual hive inspections.
Mid-2026: Prototype v0.4
Iterate based on field test results. Refinements to enclosure, sensor placement, power system, and firmware. Goal is a more robust unit that can run autonomously for months.
Late 2026: Wider pilot program
If field tests go well, expand to 10-15 hives with early collaborators. Focus on multi-hive monitoring, automated alerts, and dashboard usability.
Challenges I'm still figuring out
Sensor placement: Where you put sensors inside a hive matters enormously. Too close to the brood, and readings are skewed. Too far, and you miss critical data. Finding the optimal placement is trial and error.
Acoustic noise filtering: Bees are loud. Wind is loud. Hive vibrations add noise. Extracting meaningful acoustic patterns from all that requires sophisticated signal processing—and I'm still learning as I go.
False positives: An alert system that cries wolf too often becomes useless. Balancing sensitivity (catching real problems early) with specificity (not flagging every minor fluctuation) is tricky.
Real-world durability: Lab tests are one thing. Surviving months inside a working hive with humidity, propolis, temperature swings, and bee interference is another. January will reveal what breaks.
Cost vs. features: Building something useful but also affordable enough for small-scale beekeepers is a constant balancing act. Every sensor, every feature adds cost. Deciding what's essential vs. nice-to-have is hard.
Want to follow along?
I send occasional updates when there's real progress to share—field test results, prototype iterations, new features tested. No marketing fluff, just honest development updates from someone building this in their workshop.
For researchers or collaborators: If you're working on bee health, Varroa detection, or similar projects and want to share data, collaborate, or test prototypes—reach out. Always keen to connect with people working on related problems.
BeeSight Built in Adelaide, South Australia
Last updated: December 2024