Late-season woody invasive — Lake County, MN research flight
Pacific Northwest · invasive-plant review
Find invasive plants
from the sky.
Built in Moscow for county weed programs, conservation districts, and land managers who treat plants. Upload an orthomosaic, review detections on the map or in the field, and leave with a treatment-planning map your crew can walk.
Free, on imagery you have already flown. No quote first.
When this comes up
Nobody goes looking for a detection model.
Land managers call when a season, a contract, or a report puts them on a deadline they cannot walk fast enough to meet. These are the moments aerial survey actually earns its place.
-
Scoping a treatment contract
You need to know roughly where the infestation is and how big before you can write a scope or ask for a number.
-
Closing out a treatment season
Crews worked the site. Now you need something more defensible than "it looks better" to put in front of a board or a funder.
-
Reporting to a grant
The award requires acres surveyed, acres treated, and evidence of change over time — on a reporting calendar, not a field calendar.
-
A new report on ground you rarely visit
Someone flags an infestation on a parcel nobody has walked in years, and you need to know whether it warrants a trip.
How it works
Three steps from raw orthomosaic to validated detections.
-
01
Upload
Drag and drop multi-gigabyte GeoTIFF orthomosaics. Multipart S3 uploads run in the background with resumable progress; metadata is captured automatically from the source raster.
-
02
Detect
We tile the imagery for fast web rendering and run RF‑DETR inference at the appropriate zoom. Every detection is stamped with full provenance — model, version, run id — so any result can be traced back to what produced it.
-
03
Validate
Review on the desktop map for sweeping audits, or use offline Field Mode on a phone for ground‑truth walks. Validations stay separate from source — what the model said, and what a human said about it, both preserved.
Treatment & outcomes
From validated detections to treatment maps — and back again.
Verified findings become treatment-ready polygons and crew handoffs. After removal or treatment, re-fly the same ground. Aerial change on the stems visible from above, plus the paired field transects, gives a manager a recurrence number they can put in a report.
For under-canopy targets like forest-interior buckthorn, baseline flights are best timed to the late-fall leaf-retention window, when buckthorn stays green after native leaf-drop.
See it on your ground
A trial run on a flight you have already done.
Send us one orthomosaic from a place you know. We process it, load the results into DataHub, and hand you the review queue plus a written read. You see the work on your acres, then decide whether a season together is worth it.
- It costs nothing. No quote first, no commitment, no card. The point is to find out whether this is useful on your ground before money is involved.
- Two business days to a reply, ten to results. You get your imagery as map tiles, model detections on top of it, and a DataHub login to work the review queue yourself.
- Your imagery stays yours. We process it to produce your results, we do not resell project data, and we will not train a model on it without asking you in writing first.
- A written read at the end. Not just pins on a map — what the model found on your ground, how it lines up with what you already know is there, and what running this across your whole program would take.
Common questions
The things people actually ask first.
- How accurate is it?
- Accurate enough to change where your crews walk — which is the job. It runs as a review queue, not an oracle: the detector proposes candidates and a person decides, on the desktop map or in the field, so nothing lands in your records unreviewed. The figure worth having is how it does on your imagery rather than on ours, so we run it over a flight you have already done and you judge it against ground you know. Methods and results are on the research page.
- Does this replace field crews?
- It changes where they spend the day. The model narrows a large area to a ranked list of places worth walking, so the crew arrives with a route instead of a search. Every engagement pairs aerial candidates with field transects — the walk is what settles it.
- Can it see under the canopy?
- A drone sees the canopy surface, so we aim aerial detection where buckthorn is actually visible: edges, gaps, and openings, flown in the late-fall leaf-retention window when it stays green after the natives drop. That window is the whole trick — it turns an under-canopy problem into an aerial one across a surprising amount of a site. Interior stems are picked up on the paired field transects.
- What imagery do you need?
- A georeferenced GeoTIFF orthomosaic. Multi-gigabyte files are expected — uploads are multipart and resumable. If you have not flown the site yet, capture can be coordinated separately, and the timing of the flight usually matters more than the camera.
- What do I actually get at the end?
- Validated detections on an interactive map, treatment-ready polygons, and GIS exports you can take into ArcGIS or QGIS. Every detection carries its provenance, and every human verdict is stored separately from what the model said — so a reviewer can always tell the difference between a prediction and a decision.
- Who owns the data?
- You do. Project imagery and validation data stay under the landowner's or grantee's control as defined in the agreement. We process it under contract to produce your deliverables, and we do not sell project landowner data as a product without written agreement.