In development · Early access open
Pitchfork
See plant stress before it shows.
Four cameras. One trigger.
A handheld field instrument for plant scientists, breeders and agronomists running trials. One pull of the trigger captures colour, red, near-infrared and radiometric thermal images of the same plant. Pitchfork finds the leaf pixels, computes NDVI and canopy temperature on leaves alone, and combines them into one stress map.
Why it exists
Stress shows in temperature before it shows in colour.
A plant short of water closes its stomata, stops cooling itself and warms up, days before a leaf wilts or yellows. A thermal camera sees that. On its own, though, it also sees warm soil, sunlit stakes and plot labels, and in a sparse canopy those dominate the average. Pitchfork measures temperature and NDVI on leaf pixels only, at 0.3 to 1 m, where individual leaves are resolved and nothing but the plant is counted.
Colour
Frames the shot, checks focus and documents what the plant looked like. The picture a human reviews.
Red
Chlorophyll absorbs red light. A global-shutter monochrome camera behind a narrow band-pass filter measures how much.
Near-infrared
Healthy leaf tissue reflects near-infrared strongly. Red and near-infrared together give NDVI, per pixel.
Thermal
A radiometric thermal core reads leaf temperature directly, in degrees, not as a picture that only looks warm.
How a capture becomes a number
From trigger to stress map.
-
1
Aim and pull
A live view on the 5-inch screen cycles through the bands. Frame the plant from 0.3 to 1 m and pull the trigger.
-
2
Four images, one pull
Colour, red, near-infrared and thermal are captured on the same trigger. The red and near-infrared cameras are registered to each other with a calibration fitted on a flat target.
-
3
Leaves only
A per-pixel leaf model separates leaf from soil, stakes and labels. It can be retrained for a new crop in the companion app.
-
4
Numbers you can check
NDVI and canopy temperature on leaf pixels, combined into a stress map. Raw bands are saved as NumPy arrays with JSON metadata, so every number can be recomputed in Python or R.
Any index
Swap the filters, choose the index.
Pitchfork's two band cameras sit behind band-pass filters that swap out. Red and near-infrared give NDVI; a red-edge or green filter gives NDRE or GNDVI instead. In principle, any index built from two bands works. Pair each filtered camera with a channel from the colour camera, and one capture can even carry two indices at once.
NDVI as standard.
Red and near-infrared: the index most trials already report, measured on leaf pixels only.
Your bands.
Fit the filters your question needs. Which camera carries which band is a setting on the device, not a rebuild.
Two at once.
Each filtered camera paired with a red, green or blue channel from the colour camera: two indices from one pull of the trigger.
What's inside
Four cameras and a computer, in one hand.
Two global-shutter cameras behind band-pass filters, a radiometric thermal core and a colour camera share one black sensor plate. Behind it sit a single-board computer, a sensor hub, a battery and a 5-inch screen, all inside a light shell with a pistol grip and an orange trigger.
Your data stays yours
Hayloft: the companion app with no server.
Captures come off the device on a USB stick and open in Hayloft, a companion app that runs entirely in your browser. Every file is decoded and every map drawn on your own computer. Nothing is uploaded, there is no account, and it works offline. Leaf models trained in Hayloft and changed settings go back to the device on the same stick, checked before they are applied.
No cloud in the loop.
No upload, no account, no licence server. Your trial data never leaves the machines you already control.
Open formats.
NumPy arrays and JSON, not a proprietary container. Your statistician can read them without asking anyone.
Same maths, twice.
Hayloft recomputes captures with the device's own processing and is tested against real captures pixel for pixel, so the two never quietly disagree.
Where it stands
Proven on the bench. Next: the field.
Status as of September 2026. The core system runs on real hardware: all four cameras, synchronised capture and live processing on the device. Production optics, a field-ready enclosure, environmental sensing and validation on real trials come next. No date is promised.
- Four-camera imaging head
- Synchronised capture
- Band registration
- Real-time NDVI
- Companion software
- Production optics
- Field-ready enclosure
- Environmental sensing
- Field validation
- Refinement
- Early-access units
Early access
The first units are few.
They go to a small number of trial teams who will use Pitchfork on real plants and say plainly what works and what does not. Terms, timing and hand-over are agreed with each team. Hand-over happens in person, so the first season favours trials in the Netherlands and nearby.
Who it suits.
Plant scientists, breeders and agronomists who want leaf-level NDVI and canopy temperature on individual plants or small plots, in the field or the greenhouse.
Help shape it.
Early teams get a say in what Pitchfork becomes: which bands it measures, what the screen shows, how the data comes off. Direct support from the engineer who built it.
What is asked.
Use it on real trials, report what breaks, and share honest feedback on what the numbers were worth to you.
Get on the list
Send a short email with your organisation, crop and type of trial, where it runs and which season you have in mind. Registered teams hear first when units are ready. Registering commits you to nothing.
Planned specification
The short version.
Design targets for the first units. Subject to change as testing goes on.
- Bands
- Colour, red, near-infrared, radiometric thermal
- Band cameras
- Two global-shutter monochrome sensors behind swappable band-pass filters
- Indices
- NDVI as standard; in principle any two-band index, and two at once with colour channels
- Thermal
- FLIR Lepton 3.5 radiometric core, 160 × 120
- Working distance
- 0.3 to 1 m
- Form
- Handheld, pistol grip, trigger
- Screen
- 5-inch touchscreen with sun hood
- Battery
- About 3 to 4 hours
- Output
- Raw bands as NumPy arrays plus JSON metadata
- Software
- Hayloft, browser-based, offline, no upload
Questions
Asked so far.
When will it be available?
When it is ready for a real trial season, and not before. Pitchfork works on the bench today, but the optics, the enclosure, the last sensors and field testing all lie ahead, along with plenty of smaller things. No date is promised. Registered teams hear first.
What will it cost?
That is not settled yet. Terms for the first units are agreed with each early-access team, and what the first season teaches shapes the rest. Registering commits you to nothing.
Can a unit come to my trial abroad?
The first season relies on an in-person hand-over and quick support, so trials in the Netherlands and nearby come first. If yours runs further away, register anyway and say where.
Does it replace a drone?
No. A drone covers whole fields from above. Pitchfork looks at individual plants from 0.3 to 1 m, close enough to separate leaves from soil. They answer different questions and work well together.
Can it measure other bands?
Yes. The band-pass filters are swappable and the band choice is a setting on the device, so in principle any two-band index works, and pairing each filtered camera with a colour channel gives two at once. If your trial needs different bands, say so when you register: that is exactly the kind of input early teams give.
Where does my data go?
Onto the device, then onto your USB stick, then into your browser. Nothing is uploaded anywhere, including to Haystack Imaging.