Scouting Colorado potato beetles with drones: what a camera can see
Walking a field to count beetles is slow and samples only a few plants. Drones promise to look at every plant. How well that works comes down to one number: how many millimetres of field fall on one pixel.
Millimetres per pixel decide everything
An adult Colorado potato beetle is about 10 mm long, a young larva only a few millimetres. In photos resampled to different resolutions, the picture is clear:
| field per pixel | what you see |
|---|---|
| 0.5 mm | a clear beetle; larvae about 20 pixels long |
| 1 mm | a striped blob: enough to flag the plant |
| 2 mm | a dot |
| 4 mm | gone |
Detection models confirm it: objects under 32 × 32 pixels count as "tiny" in computer vision, under 16 × 16 as "very tiny", and accuracy drops quickly below that.
What the research shows
Ghent University flew a drone at 10 m with a long lens over a Flemish potato field, at about a third of a millimetre per pixel. Their detector found 53 % of the beetles (recall 0.53), and 70 % of what it flagged was really a beetle (precision 0.70). That is good enough to direct a closer look, not to count every insect.
Some beetles cannot be seen from above at all. Egg masses sit under the leaves (85 to 92 % of them), and adults and larvae move under the leaflets in the heat of the day. Field studies put the share of adults visible from above at about 85 to 90 %, and of large larvae at about 80 % or more.
Height, sharpness and time
To reach 1 mm per pixel, a small camera drone of the DJI Mini class (about 12 megapixels of real detail) has to fly about 3 m high; each photo then covers about 4 × 3 m. Scouting one hectare that way takes about 34 minutes. Sharper cameras fly higher and finish faster:
| scout camera | price | min per ha |
|---|---|---|
| small camera drone, 12 MP real | €300 | 34 |
| DJI Mavic 3 Enterprise, 20 MP | €3,600 | 21 |
| 45 MP survey camera (DJI Zenmuse P1 on a Matrice) | €17,000 | 11 |
A caution on megapixels: small drones sold as 48 MP or more often use quad-Bayer sensors that resolve far fewer true megapixels. It is the real detail that counts.
What the scout camera sees is one of the narrated, interactive chapters on photonicinsecticides.com: 3D scenes, real drone footage and every number traced to the study it comes from.
Cheap and many beats sharp and expensive
Because the scout flies all season, every two days, its cost per hectare depends on its price and how many hectares it can cover. In our models, many cheap camera drones cost about €32 per hectare per year for scouting; a fleet of 20 MP enterprise drones about €56; a 45 MP survey camera on a six-kilo drone about €109. A cheap drone with a sharper sensor would be the real win: with 45 MP of real detail, the whole service would drop from about €51 to €29 per hectare per year.
From photos to targets
Every photo is geotagged, so each detected beetle becomes a GPS coordinate. That list is what a hunting drone needs: it flies to each spot, looks close up with its own camera, and decides there. Scouting does not have to be perfect, because the hunter checks at close range, and patrols every two days catch what was hidden last time. If you already have a scouting heat map, you can price a laser-drone pass with it: see from heat map to cost per hectare.
Sources
- Wieme et al. (2023). Efficient detection of Colorado potato beetles in ultra-high-resolution UAV imaging of potato fields. CVPPA 2023 (Ghent University).
- Potato_Beetle_Detection dataset (Roboflow Universe): 1,824 field-camera images with beetle and larva boxes.
- Small-object detection in pest images, Frontiers in Plant Science (2024): definitions of tiny objects.
- Gibson et al. (1925): egg masses under the leaflets; Zehnder (1990): upper-canopy counts.
- DJI specifications (Mini, Mavic 3 Enterprise, Zenmuse P1); Pix-Pro field tests of quad-Bayer sensors.
- Scouting time and cost: photonicinsecticides.com scout camera model (llms.txt).