Industrial AI
Industrial AI means models that run in the field on real measurement data, not in a chat window. An off-the-shelf AI API does not know your orchard, your line or your map. We train the model on your data, measure it on sites it never saw in training and turn it into a working application: object detection in drone and satellite imagery, CAD drawings from orthophotos, prediction on spectral and sensor data, and LLM assistants that run on your own servers.
Request Detailed Quote
Proven in the Field
PlantFarm
Drone-based tree analysis with GETEK
Model Types
Vision · Signal · Language
Detection, segmentation, regression, RAG
Runs On
Server · Cloud · Edge
Your data can stay on site
All Features
Request Detailed Quote
Contact us to get detailed information about our products and receive a custom price quote.
From the Field
Every method described here has been used in a real project. The numbers are those projects' own measurements.
PlantFarm Scan: Tree Detection from Drone Imagery
PlantFarm Scan, built with GETEK, finds the crown of every walnut and almond tree in a multispectral drone orthophoto. We trained a YOLOv8-seg model on flight images we labelled ourselves.
Tiled inference
The orthophoto is scanned in 640-pixel tiles with 25% overlap; crowns cut at tile edges are stitched back together.
Measured on the crown mask
NDVI and NDRE are computed over the tree's actual crown mask, not its bounding box.
Persistent tree ID
The same tree keeps the same ID across flights, so its course over the season can be followed.
Multispectral Plant Health and Trends
Red-edge and near-infrared bands change before leaves turn yellow. PlantFarm keeps a per-flight index series for every tree and surfaces the ones getting worse across the project.
9 vegetation indices
NDVI, NDRE, GNDVI, GCI, RCI, plus VARI, TGI, ExG and GLI for RGB-only cameras.
Ground classification
Soil, dry grass, green grass, healthy tree, stressed tree and shadow; each tree is compared with its surroundings.
Regional reports
PDF, Excel and CSV reports for an area drawn on the map and the selected flights.
Milk Analysis from NIR Spectra
Our model, trained on open data, predicts the fat and protein content of raw milk from its NIR spectrum. We published the model and an in-browser simulator on Hugging Face.
Time-split test
Trained on the first six weeks, tested on the last two: fat R² 0.979, protein R² 0.965.
Classical first
PLS and a 1D-CNN compared with the same preprocessing; whichever wins is used.
Open source
The model and simulator are open for anyone to try.
An AI Assistant That Reads Your Measurements
PlantFarm's AI Operator answers questions using the project's own trend data together with walnut disease literature. It runs on a local LLM, so data never leaves the site.
Measurement context
The model reads from the same services as the map, so the screen and the assistant never disagree on numbers.
Curated corpus
The literature passages closest to the question are found with multilingual embeddings.
On-premise
Runs on a single-GPU server with Ollama.
More Information About This Solution?
Our technical team is ready to provide custom integration solutions for you