Perceive · Folira

See every plant, in real time

Folira is Cropite's edge perception engine. It runs on Jetson, reads multiple cameras on a moving machine and labels every plant: crop or weed, pest damage, disease, nutrient stress and ripeness.

Folira · frame 08412, camera 3Example
  • plant 455 · soybean · 0.97Healthy
  • plant 457 · amaranthus · 0.94Weed
  • plant 458 · soybean · leaf spot 0.71Disease
  • plant 459 · soybean · 0.96Healthy
  • unknown object · 0.41Send to label

ExampleClasses and confidences are illustrative.

Capabilities

What Folira does

Detection

Every plant, every frame

Detection and instance segmentation of crop and weed plants at implement speed.

Pipelines

Many cameras, one unit

DeepStream pipelines with on-device tracking across cameras and frames.

Model packs

Per crop and region

Packs tuned to a crop, its weeds and its soils, updated over the air.

Uncertainty

Knows when it is unsure

Confidence and novelty scores flag plants it has not seen for labelling.

Output

One event per plant

A structured detection record that Targis, Acreon and Agrona read.

Conditions

Built for outdoors

Training covers dust, glare, low sun and plants that overlap.

Detection classes

Filter the planned detection classes

The first model pack is planned for one row crop and its common weeds. Everything else is on the draft roadmap.

Class
Crop

Detection classKindCropsModel pack
SoybeanCropsoybeanFirst pack (planned)
CornCropcornFirst pack (planned)
LettuceCroplettuceLater (planned)
GrapevineCropgrapeLater (planned)
Amaranthus (pigweed)Weedsoybean, cornFirst pack (planned)
Grass weedsWeedsoybean, cornFirst pack (planned)
Weeds in lettuce bedsWeedlettuceLater (planned)
Leaf spotDiseasesoybeanLater (planned)
Downy mildewDiseasegrape, lettuceLater (planned)
Insect feeding damagePestsoybean, cornLater (planned)
Nitrogen stressStresscornLater (planned)
Ripe bunchRipenessgrapeLater (planned)

How it works

Folira in four steps

1

Capture

Cameras on the boom stream frames to the edge unit.

2

Detect

TensorRT models find and label every plant.

3

Track

Each plant gets an ID that survives from frame to frame.

4

Emit

A detection event goes to Targis in milliseconds.

Specification

Planned specification

Planned values are design targets for the first release and will change with pilots.
ItemPlanned
Edge hardwareJetson Orin or IGX
Cameras per unitUp to 8 (target)
Frame to detection<30 ms (target)
ClassesCrop, weed, pest, disease, stress, ripeness
Model deliveryOver the air, per machine group
OutputJSON events, as-applied maps

Status

Pre-launch

Nothing on this page is shipping today. Items marked planned are on the draft roadmap.

Draft roadmap

Place in the loop

Where Folira sits

Folira is where the loop starts. Its detections drive Targis's actions, fill Acreon's twin and tell Agrona where to send the next scouting pass. Plants it is unsure about go to labelling, and Synora makes more examples like them.

Questions

Questions about Folira

An agronomist or an engineer reads every message, not a sales queue.

Which cameras does Folira need?

Industrial RGB cameras mounted over the rows. The kit and layout depend on the machine; we specify them per setup.

Can it work at night?

With lighting on the boom, yes. Night passes are planned after daytime packs are proven.

Does it identify weed species?

Groups first (broadleaf and grass), then species where it changes the treatment.

Jump to a page or section