Yield forecasting
On this page 11
Count what is actually out there
Drones count plants, fruit or ears and turn the counts into a yield estimate for each part of the field.
| Category | Detect |
| Slug | yield-forecasting |
| Cadence | Once at establishment, then at two or three points through grain fill or fruit set |
| Payload | High resolution RGB, multispectral, RTK |
| Needs a visit first | No |
The problem
Harvest planning, storage booking and forward selling all run on an estimate, and the estimate usually comes from a hand count in a few square metres extrapolated across a hundred hectares. A bad estimate books the wrong number of trailers and sells grain that is not there.
How it works
| Step | What happens |
|---|---|
| Fly low | A high resolution pass captures individual plants, ears or fruit at a density the model can count |
| Count | Objects are counted per square metre and the counts are aggregated by zone rather than averaged flat |
| Model | Counts are combined with your variety, row spacing and historical yields to produce an estimate range |
| Update | The forecast is re-run on each later flight, so the range narrows as harvest approaches |
What you get
- Plant or fruit count per zone
- Yield estimate with a range
- Establishment gap map
- Harvest sequencing suggestion
What the dashboard measures
- Counted objects per square metre
- Estimate range width
- Gap area as a share of the field
- Estimate against final weighbridge
The last one is the only honest measure of this capability, and it arrives once a year. Record it every year without exception: it is what makes next season's range credible.
Where it matters most
Winter wheat, maize, orchards, berries, cooperatives.
Build it
Two different products under one name
Be clear about which one you are building. They have different economics.
| Counting | Estimating | |
|---|---|---|
| Question | How many plants, ears, fruit are here | How many tonnes will come off |
| Method | Object detection over high resolution frames | A model relating counts to weight |
| Accuracy | Good, and verifiable immediately | Depends on a season of calibration |
| Sells to | Anyone deciding to redrill, replant or book pickers | Anyone selling forward |
Counting is buildable in a season. Estimating needs a calibration set of counts paired with weighbridge tickets, which is a second season. Ship counting, and be honest that estimates start wide.
Flight parameters
| Crop | Target | Altitude |
|---|---|---|
| Cereal establishment | Plants per m² | 10 to 15 m, 0.3 to 0.5 cm/px |
| Cereal ear counting | Ears per m² | 5 to 10 m, or a ground rig |
| Maize establishment | Plants per row metre | 20 to 30 m, 0.7 cm/px |
| Orchard fruit | Fruit per tree | 8 to 15 m, oblique passes both sides of the row |
| Berries | Fruit per plant, per flush | 5 to 10 m under tunnel headroom |
Note the altitudes: ear counting is at the edge of what a multirotor can do safely over a standing crop, and orchard counting wants oblique views rather than nadir, because the fruit is on the sides of the canopy. Both push you toward a smaller aircraft flying transects rather than a survey grid.
Hardware
| Item | Why | EU source | US source | Indicative |
|---|---|---|---|---|
| High resolution RGB, global shutter | Counting at 0.3 cm/px is unforgiving of motion blur | Sony ILX-LR1 | B&H | €2,500 to €3,200 |
| Compact alternative | 12 MP, tiny, good enough for maize and orchards | Raspberry Pi GS camera at BerryBase | Adafruit | €55 plus lens |
| Small airframe | Low, slow, close to the canopy | Holybro X500 V2 | GetFPV | €400 to €600 |
| Precision rangefinder | Constant AGL is what keeps the count per m² honest | LightWare SF000/B | RobotShop | €200 to €300 |
| Calibration frames | Quadrats hand counted on the ground, the ground truth for everything | wire and paint | same | negligible |
| Gimbal for oblique work | Orchard and vineyard counting needs a side view | Gremsy or a fixed 45 degree mount | GetFPV | €0 to €900 |
In-house software
| Stage | What we run | Licence | Replaces |
|---|---|---|---|
| Detection or density | Point-supervised counting (density map regression) rather than boxes | own code, PyTorch BSD | commercial counting services |
| Tiled inference | SAHI over full-resolution frames | MIT | |
| Overlap de-duplication | Project counts to ground coordinates, deduplicate in the overlap | own code | |
| Aggregation | Counts per zone using the zone layer from fertilisation | own code | |
| Estimation | Regression from counts and indices to weighbridge tonnes, per crop and variety | scikit-learn, BSD | |
| Range | Quantile regression or a bootstrap over the calibration set. Publish an interval, never a point | own code |
Density map regression is the right choice for dense objects like wheat ears. Labelling a box around each of 400 ears in a frame is unaffordable; clicking a dot on each is not, and the dot is all a density model needs.
Cost efficiency
- De-duplicate, or you will double count. At 70% overlap, every object appears in three frames. This bug is easy to ship and hard to notice, because the number is merely wrong rather than obviously broken. Validate against hand-counted quadrats every single season.
- Publish a range, always. A point estimate that is 12% out destroys trust in the whole platform. A range that contains the answer builds it. This is a product decision that costs nothing.
- Hand counts are your cheapest asset. Twenty quadrats per field costs an afternoon and calibrates everything. Build the capture form first.
- Sell the decision, not the number. Nobody buys "an estimate". They buy "book six trailers instead of eight", "redrill these 3.1 ha", "the pickers come Thursday".