Why we built Noktura
Reinventing the recipe book for every new agtech idea does not scale. Yet, it seems this approach is the modus operandi of the industry. Every new crop x weed x environmental combination requires companies to collect their own data, train their own models and overcome the same bugs as those before them. It is incredibly inefficient. So, Noktura is the recipe book. We give farmers and researchers the tools to build their own AI ecosystems with shared hardware, software, and data, so progress made in one paddock or on one farm benefits everyone.
You get to choose who you share your data with, it can be fully open or locally open.
The Data Challenge in Agriculture
Modern computer vision models for agriculture (e.g. weed detection, crop health monitoring, yield prediction) require vast amounts of diverse training data. Deep learning models are good, but still are challenged by generalising into new fields. Plus, agricultural imagery remains fragmented across research groups, farmers, companies and farming groups, with most data never leaving the institutions that collected it.
The result is duplicated collection efforts creating inefficient deployments and worse model performance, models that only work in narrow conditions, and slower progress for the entire field. Companies are reluctant to enter new crops, countries or conditions if the scale isn't there. So, why not provide the infra instead?
Noktura is the infrastructure for vision AI in agriculture.
Enabling locally open data
We enable frictionless data sharing at different settings and allow groups to work together on locally open models. Sharing gets easier, models get better, and efficiency improves for everyone.
Private
Your data, your control. Keep datasets private while still benefiting from organisation and metadata tools.
Group Shared
Share with trusted collaborators (your research group, institution, or consortium partners) without making data fully public.
Public
Make datasets available to the entire research community. Get credit for your contributions while accelerating the field.
AgContext: The Metadata Standard
Raw images are only as useful as their documentation. A photo of a field without context (what crop, what growth stage, where, when, under what conditions) has limited value for training robust models.
Noktura implements AgContext, a standardised metadata schema for agricultural imagery. Every dataset uploaded captures:
Crop Information
Species, variety, BBCH growth stage.
Location and Time
GPS coordinates, capture date, timezone.
Environmental Context
Weather conditions, soil type, illumination.
Capture Equipment
Camera, lens, platform (drone, tractor, handheld).
This structured metadata enables powerful filtering (find all images of wheat at BBCH 30-39 captured under cloudy conditions) and helps researchers understand exactly what conditions a model has been trained on.
Who Benefits from an Open Ecosystem
Academic Researchers
Validate against shared datasets from different regions instead of running expensive solo field campaigns. Build on reproducible benchmarks. Get citations when others use your shared work.
AgTech Companies
Advance faster when you are not building everything from scratch every time. Test against diverse public conditions before deployment. Contribute back to build credibility in the agricultural community.
Farmers and Agronomists
Contribute imagery, models, and hardware modifications from your fields. The tools you use get better when they are built around real conditions, not lab conditions.
Global Agtech
The whole field moves faster and more efficiently when not every company is starting from zero. User-centred tools, flexible enough to fit niches, build on each other rather than fragmenting effort.