From a palm to a decision, once a day
The pilot replaces monthly manual trap checks with cameras on host palms and a small amount of software. Nothing here runs in real time: photos are collected overnight, processed each morning, and the dashboard is rebuilt. That keeps the system cheap, inspectable, and free of any server that could fail in the field.
Capture
Cellular trail cameras mounted below the crown of a host palm photograph it at night, when adult coconut rhinoceros beetles are active, and send thumbnails over LTE.
Collect
Each morning the photos and their weather records (wind, temperature, moon phase) are pulled from the camera vendor's cloud into the pilot's own database. Every image is kept.
Screen
A detector trained to recognise adult CRB scores every frame and draws a box where it thinks a beetle is. Most frames are wind-blown fronds and score near zero.
Review
A person looks only at the handful of frames the detector flags, typically under two per night, and confirms or rejects each with one keystroke.
Report
The dashboard is rebuilt with confirmed detections, camera health and the green-waste layers. A public view withholds imagery and parcel-level detail.
What a detection looks like
What is real in this demonstration
- All photo counts, capture times, wind and weather values, battery levels and camera health flags come from the three cameras installed on August 27, 2026.
- The map, the target areas from the pilot plan, and the camera positions (rounded) are real.
- The detector is a real model, fine-tuned from the University of Hawaiʻi CRB-Cam detector, and the scores shown are its real outputs.
What is not
- Every "confirmed" detection in this build is a synthetic example, labelled as such wherever it appears. No adult CRB has been confirmed on camera at Ko Olina yet.
- The green-waste pile and damage inventories are empty until the drone flight is annotated.
- Real photographs of flagged frames are withheld from public views; only the synthetic examples carry imagery here.
Why the cameras have not caught a beetle yet, and what changes next
Three weeks of data showed that motion triggers are dominated by wind, that the daily capture spike at 17:00 to 19:00 is sunlight on the crown, and that only one frame in six falls in the night window when beetles fly. University of Hawaiʻi trap studies place two thirds of beetle activity within three hours after dusk. The next step is to move the cameras closer to the crown, restrict capture to the night window, and add a scheduled capture stream, then let the detector and reviewers do their work. The full deployment guidance and the classifier design are in the Phase 0 technical report.