Automated roadside video surveys for detecting and monitoring coconut rhinoceros beetle damage to coconut palms
Aubrey Moore University of Guam
Trevor Jackson AgResearch New Zealand
CRB Action Group Webinar | December 9, 2020
https://aubreymoore.github.io/crb-roadside-slides
We need standardized surveys of CRB damage
- to monitor changes over time and space, especially in response to control activities
- for early detection of CRB in new geographic areas
Jackson 5-level CRB damage index
| 0 | Zero | No CRB damage symptoms evident |
Jackson 5-level CRB damage index
| 1 | Light | Notching or tip damage. <20% foliar loss. |
Jackson 5-level CRB damage index
| 2 | Medium | Multiple fronds affected. Notching and breakage. 20%-50% frond loss. |
Jackson 5-level CRB damage index
| 3 | High | Multpile fronds affected. Notching and breakage. >50% frond loss.<20% foliar loss. |
Jackson 5-level CRB damage index
| 4 | Non-recoverable | Palm dead with growing point destroyed. |
Data Acquisition Options
- direct observations
- still images
- videos
Advantages of using videos over still images
- Rapid acquisition of continuous data over large areas
- Individual palms are viewed multiple times, from multiple angles
Data acquisition
- OpenCamera app
- GPSLogger app
Data processing
Evaluation of Survey Results
100 of 175,000 tree objects selected randomly
Evaluation of Survey Results
100 selected tree objects extracted from videos
Evaluation of Survey Results
100 selected tree objects classified by human
Rules for for Combining Results from Object Detectors
| TreeObj Damage Index | VcutObj Count | Modified Damage Index |
|---|---|---|
| 0 | >0 | 1 |
| 1,2 | 0 | 0 |
Evaluation of Survey Results
After modifying damage indices using v-cut counts
Future Prospects
Acknowledgments
Thanks to:- Don Scott, Savan Visalpara, and Rush Tehrani at OnePanel Incorporated for developing the object detectors and workflow.
- Christian Cayanan for doing the surveys.