When standard
vision software
hits its limits.
Products
Vega
industrial imaging software
DetailsVega processes your image data, evaluates it and produces the needed result — a report, a sorting signal, an alarm. You can use it locally, on servers and in the cloud.
- Real-time processing with short, predictable latencies
- High throughput on modest hardware
- Can also process hyperspectral and 3D data
- AI and classic algorithms can be combined
- Easily connects with different systems
Phoenix
AI model training
DetailsPhoenix turns your images into neural networks. You select the use case, Phoenix automatically handles dataset management, training and testing.
- Can run on your own hardware — your images stay in house
- Covers more than 30 use cases (including hyperspectral models)
- We can customize model architecture for your needs
Custom Software
our software,
extended for your taskDetailsNot every task is covered by what exists. When something is missing — a particular evaluation, a machine interface, a measurement specific to your process — we build it for you into our existing software.
- Feasibility checked on your images first
- AI models trained for your objects and defects
- Custom-made evaluations and interfaces
- Maintained with every release
- Built on established software, not from scratch
KESTRELEYE in numbers
2016
Founded
10+
Industry partners
8
Industry sectors
13
Countries
2
Locations
25+
Applications
123,456,789
Processed data in GB
Vega
industrial imaging software
Vega is the software that does the actual work in your application: it takes in image data, processes it, and produces a result the machine can act on — a sorting signal, an alarm, a measurement, a report. It is used in production lines, on servers and in the cloud.
Ready to use for three tasks
- Sorting — Vega Sort drives band and free-fall sorters. It decides per object and delivers the ejection signal in time for the object to still be in the right place — latencies below one millisecond are achievable, which is the part that decides whether a sorter works. Where colour alone is not enough — different materials that look alike, contamination, ripeness, moisture — hyperspectral data tells them apart.
- Continuous monitoring — Vega Guard watches a running process without a trigger: endless material, changing scenes, high data rates. It reports what deviates, from a known defect to something simply not seen before.
- Inspection — Vega Inspect checks single parts on a trigger: present or missing, correct or faulty, within tolerance or not. Many suppliers cover this; customers usually come to us with the cases that did not work elsewhere.
- Anything else — the same functions can be put together into a bundle for your task, without development work on our side.
▾What Vega works with?
Monochrome and colour images, hyperspectral data and 3D point clouds — in the same pipeline, with AI models and classic algorithms side by side. Whatever your sensor delivers, the result is one decision, not three separate systems.
▾Where it runs?
In the line on a local machine, on your server, or in the cloud. Without a camera it processes stored image data in large volumes — the same software, the same results, a different setting.
▾What comes with it?
Vega talks to your control system, so the result arrives where the decision is executed. It records raw and processed data on request, which is what you need when a customer complaint has to be traced back months later.
Explore the technical details in our datasheet
Download PDFPhoenix
AI model training
Phoenix is where the AI models come from. You bring images from your process, choose what the model should do, and Phoenix takes care of the rest — preparing the data, training, testing. What comes out is a model that runs in Vega, plus a report showing how well it performs.
What you can have trained
- Sorting and grading — which type, which grade, good or reject
- Finding and outlining — where something is, how large, how many
- Measuring — thickness, height, moisture, ripeness, quantity: a number instead of a verdict
- Spotting the unexpected — deviations without having to collect examples of every possible defect first
- Material and content — what something is made of, read from hyperspectral data
Every one of these works with monochrome, colour and hyperspectral images. The last two are where standard tools usually stop.
▾Who operates it?
Your application engineers, not a data science team. The graphical interface asks for the few decisions that matter; command line and API are there for whoever wants to automate the process.
▾Where your images stay?
Phoenix runs on your own hardware if you want it to — image data from your process never has to leave the building. Cloud is available where that is the more practical route.
▾What you get back?
A trained model ready to run in Vega, and a report with the numbers behind it: how reliably the model performs, where it is weak, what the training was based on. That report is what you hand to quality management when someone asks how the decision is made.
Explore the technical details in our datasheet
Download PDFCustom Software
our software,
extended for your task
Not every task is covered by what exists. When something is missing — a particular evaluation, an interface to a machine nobody else runs, a measurement specific to your process — we build it into Vega and hand over a system that works.
▾How it goes?
- We look at your task and your images.
- We tell you whether it is feasible — before you commit to anything.
- We build the missing part into Vega.
- We put it into operation in your line.
- It stays part of the product and is maintained with every release.
The last point is the one that matters over the years: your extension does not become an orphan that nobody dares to touch.
▾Where custom starts?
If your task can be covered by functions that already exist, you get a bundle put together for you — no development, faster, cheaper. Custom software begins where a function does not exist yet. We will tell you which of the two applies once we have seen the task.
▾What comes with it?
Feasibility studies, integration into your existing automation, training for your team, and joint development projects. These are part of how we work, not a separate product line.