Back in 2025, we shared that Elegant Media had partnered with Macquarie University — backed by Hort Innovation — on an ambitious AI spotting bug detection project: building Australia’s first AI-powered smart trap to identify Fruit Spotting Bugs and Banana Spotting Bugs.
The final report is now in, and the results confirm what we set out to prove: a purpose-built AI system can identify these pests with laboratory-grade accuracy, and the hardware to put that intelligence in a paddock is real, built, and already field-tested on a commercial Australian farm.
This is a summary of what we built, what we found, and what happens next.
The Problem: A Pest That Costs Growers Millions
Fruit Spotting Bugs and Banana Spotting Bugs are native pests that attack a wide range of Australian horticultural crops — macadamia, avocado, custard apple, lychee, passionfruit and papaya among them. In heavy infestations, fruit-spotting bug damage can wipe out more than 90% of green fruit, and in macadamia it’s a leading cause of factory rejects, costing the industry tens of millions of dollars a year.
Despite the scale of the damage, growers have historically had only one real tool: manual field scouting and hand-checked sticky traps. It’s slow, labour-intensive, and easy for pest pressure to build up unnoticed between visits.
As agronomist Jarrah Coates of Coates Horticulture in Cooroy, QLD, put it in a recent Hort Innovation update, once macadamia shells harden late in the season, fruit spotting bug populations become very difficult to monitor accurately — and the industry has long needed a better tool to close that gap.
What We Built: A Two-Stage AI Spotting Bug Detection System
The project (BY23000, funded through Hort Innovation Frontiers with co-investment from Elegant Media and the Australian Government) was delivered in two stages.
Stage 1 — The AI Spotting Bug Detection Model
Our objective was to build a specialist AI model capable of identifying spotting bugs at 80%+ accuracy. We trained a YOLO11-based object detection model on a purpose-built dataset of 1,777 farm and laboratory images, all manually labelled, using an NVIDIA A100 GPU platform.
The result, benchmarked under controlled test conditions:
- 93.8% balanced detection score (F1 score)
- 94.6% precision — when the model flags a bug, it’s right 94.6% of the time
- 93.1% recall — the model successfully finds 93.1% of bugs present in an image
- 96.2% [email protected]
- 3.1ms inference speed per image — fast enough for real-time use
To our knowledge, and confirmed by a targeted literature review during the project, this is Australia’s first specialised, benchmarked AI detection model for FSB and BSB — a capability that simply didn’t exist before this project.
Stage 2 — The Smart Trap Prototype
An accurate model is only useful if it can run somewhere. Stage 2 turned that AI capability into a physical, field-ready device: a self-contained, solar-powered, cellular-connected smart trap built around a single-board computer, paired with a high-resolution camera that photographs an OCP lure and sticky board at set intervals.
The build includes:
- A 100Wh LiFePO4 battery charged via a 12–18V solar panel
- On-device telemetry — CPU/memory usage, cellular signal, temperature, humidity, air pressure, ambient light, battery and power-rail voltage/current
- A real-time clock to keep timestamps accurate through cellular outages
- A heavy-duty metal housing engineered to mount on standard star pickets, shipped as a single-box kit
- A cloud dashboard receiving photos and telemetry for remote monitoring and AI inference
Over 50 physical component prototypes were trialled in-house at Elegant Media before arriving at the final field-ready build.
From the Lab to a Working Orchard
In February 2026, the prototype was deployed on a commercial avocado farm at Mt Garnet, in the Tablelands region of North Queensland, for a 10-week field trial — timed to coincide with the avocado season and expected BSB activity.
The trap was mounted on a star picket with the lure and sticky board positioned roughly a metre from the camera, and monitored remotely by the Elegant Media team via the cloud dashboard, with a local Project Reference Group participant visiting every 1–2 weeks to service the sticky boards and lure.
The engineering held up end-to-end. Installation required no electrical or technical expertise — just connecting the battery and flipping a switch — which farmer feedback confirmed was straightforward even for a non-technical user. The camera, solar power system, and cellular telemetry all operated as designed for the full 10-week window, with real-time environmental and system health data flowing to the dashboard throughout.
What We Learned — Being Upfront About the Trial
In the interest of transparency (as reported in the full final report): no live BSB were captured on the sticky board during the observation window, so AI spotting bug detection accuracy under live field conditions — as opposed to lab-benchmarked accuracy — remains inconclusive for now.
That’s a meaningful, and honestly useful, finding in itself. The likely contributing factors identified in the report include the farm’s fortnightly commercial spray programme suppressing pest activity near the trap, sticky boards losing adhesive effectiveness within days in field conditions, and the trap’s placement on the outer canopy rather than the interior, where the farmer observed the most spotting damage.
None of these point to a flaw in the AI or the hardware — they’re exactly the kind of real-world variables a first field trial is designed to surface. The recommendations arising from the trial are specific and actionable:
- Run multi-season trials to rule out seasonal pest activity as a factor
- Trial the new BSB lure formulation currently in development at Macquarie University alongside the existing OCP lure
- Reposition future traps closer to the tree trunk / interior canopy, where growers report higher pest activity and reduced sun exposure for the sticky boards
- Increase solar charging capacity for regions with extended overcast periods
- Reduce the overall unit weight and simplify the sticky-board swap mechanism based on direct farmer feedback
Part of a Bigger, Coordinated Effort
This project doesn’t sit in isolation. As Hort Innovation outlined in a recent update, it’s one arm of a multi-pronged, industry-funded research push against spotting bugs — running alongside genomic research at Queensland University of Technology into species-specific control targets, and lure-and-kill trap research with the Queensland Department of Primary Industries.
Our role, in partnership with the Macquarie University team led by Dr Soo Jean Park, sits at the intersection of chemical ecology and AI: Macquarie develops and refines the pheromone lures that draw pests to the trap, while Elegant Media builds the AI and hardware that identifies and counts them. It’s a project that was originally announced with a $1.49 million Hort Innovation grant back in 2024, and this final report is the first full look at what that investment has delivered.
What’s Next for AI Spotting Bug Detection
The technical foundation — the AI model and the trap hardware and software stack — is complete and validated. The recommended next phase is a multi-season, multi-site field validation program to confirm live capture and detection performance across different farms, lure placements and climates, alongside continued lure development at Macquarie University.
For an industry that’s relied on manual scouting for decades, having a benchmarked, purpose-built AI spotting bug detection capability — and a working, field-tested device to run it on — is a genuine first step toward automated, real-time pest surveillance for Australian horticulture.
Read the full technical breakdown in the official BY23000 final report on the Hort Innovation site.
The “Using artificial intelligence to develop automated monitoring systems for banana and fruit spotting bugs” project (BY23000) is funded through Hort Innovation Frontiers with co-investment Elegant Media and contributions from the Australian Government.
Have an AI or AgTech idea? Let’s work together to make it a reality.
