Previous Next

HySpex is launching a new line of industrial hyperspectral cameras: Baldur

HySpex is today the industry-leading hyperspectral manufacturer in the high-end hyperspectral market. Our focus is, and will always be, to give you pixels with exceptional spectral fidelity. We have a strategy to be fully transparent in what we do with the data when converting the data from raw format to at-sensor radiance or reflectance.

Today the hyperspectral industrial market is flooded with cameras that have a lot of spectral and spatial distortions and at the same time blurry optics per pixel. This will cause unreliability issue with the capabilities of the systems. HySpex is now offering affordable industrial systems with high data quality and with low spatial and spectral distortions.

Baldur is utilizing the same optical design as our line of classical systems. As a research tool its good to have as sharp camera spectrally as possible, but as an industrial system its good that the spectral resolution is 2 bands (2 x spectral sampling). The reason for this is that you can then reproduce the spectral frequencies in the signal (Nyquist) (read more). It's also important that the spectral resolution is not much more than 2 bands because you want as much information per data rate as possible, so an even blurrier system will give you less information per datarate. With the modification we have done to the system we are now gathering 4 times as much light through the system, making it even more light-sensitive, which is especially important for the high speed and low exposure applications.

In one of our previous article about sharp optics we defined the term Effective Pixel Count as:

"the total spatial pixels divided by the spatial resolution in pixels, you will get the number of effective pixels. If the distortion per effective pixel is large than 10% you must reduce the effective pixel count until you have distortion at 10% of your effective pixel. "

Why are low optical distortions important for industrial applications?

In a previous article, we have shown why it is important, in real applications, to have low keystone and similar PSF for different wavelengths.

In another peer-reviewed article, "Resampling in hyperspectral cameras as an alternative to correcting keystone in hardware, with focus on benefits for the optical design and data quality", it is shown that the relative errors in the spectrum per pixel as a function of keystone is as shown below.

No alt text provided for this image

An example from this graph is that 42% keystone corresponds to a 10% relative error in the spectrum. As you can understand this will have a huge impact on the quality of the instrument and repeatability of the system. Basically a system with large keystone will be unstable, the spectrum will in one pixel be highly influenced by the spectrum of the object next to your pixel, basically making the system not repeatable.

All Baldur cameras have less than 15% keystone and smile distortions per native pixel and native band, and less than 10% keystone and smile per effective pixels and effective band.

All Baldur cameras are correcting for keystone and smile in the optics, so we are not introducing an additional processing step (Resampling).

What makes Baldur line of cameras so good?


All Baldur cameras support multiple regions of interest (MOI). The Baldur V-1024N supports up to 8 independent regions and you will get a speed increase relative to the reduction of spectral bands from the native resolution. The S-384N and S-640i-N both support independent band selection from one band to the full range without being connected bands, here also you get a speed increase directly proportional to the reduction of the bands relative to the native number of bands.

All Baldur cameras have various FOV options for different working distances.


All camera can be triggered internally, and all cameras support several kinds of external triggering(TTL at different levels and LVDS). All cameras are operating in “Integrate While Read mode" making it possible to use almost the whole frame period for exposing the detector, this is very important when you will be running the system fast in the industry.


The Baldur line of industrial camera delivers the most information per data rate out of all hyperspectral cameras available today. And the speed is scalable with reducing the number of spectral channels readout.


All cameras are still delivered with a traceable calibration to NIST and PTB standard. They have a spectral resolution of two bands making it spectrally reliable when using multiple cameras with the same model/algorithm.

All Industrial camera within the same wavelength range have the same center wavelengths.

The design is originally made for airborne use and has been flying since 2003, the design is super robust and can withstand vibrations and harsh environments. HySpex has been delivering hyperspectral cameras since 2003 and has no products that have reached the end of life.

All Baldur cameras have less than 15% keystone and smile distortions per native pixel and native band, and less than 10% keystone and smile per effective pixels and effective band.

Light sensitivity

The light sensitivity of the industrial line is about 4 times higher than our classical system making it perfect for high-speed applications. 


All HySpex cameras; Baldur, Classical and Mjolnir are delivered with a high-end SDK and library that makes it simple to integrate the HySpex HW into any third-party software and hardware solution. 

Third-Party processing software

HySpex is compatible with all third-party processing software suppliers on the market today and we are open to collaborate with everybody. HySpex has started a closer collaboration with Prediktera by investing in Prediktera. We have since early 2018 been working closely with them, developing the best possible HW and SW solution on the market. Today we offer seamless HW and SW integration with all cameras and accessories. Together we offer:

  • Turnkey HW and SW solution for the customer to test and develop their own models/applications easily and fast.
  • Both High-end line of hyperspectral cameras for R&D and an affordable industrial line. We offer a special tool to downgrade the data and a tool to inform the user what wavelength range is the best for a given application.  
  • Fast cameras that can run at max speed while running the models realtime on a runtime engine that supports both CPU and GPU computing. 
  • Easy export of models from offline SW package to the runtime engine.
  • “Connector Software” that takes care of interfacing with the customer's graders, sorters or similar. This SW solution takes care of the timing of the mid-exposure of each frame (or position and time of objects) and when you need to make a decision on what to do with the results. The SW has been interfacing with several different machine suppliers’ software to control PLCs for sorting applications.

Ending remarks

All camera models by NEO not marked with the letter "N" in the name are extremely sharp per pixel and per band, and have very low distortion per pixel, thus making it a perfect research tool.

The Baldur line of cameras has focused on making the spectral resolution exactly 2 bands to make it spectrally more repeatable to avoid spectral aliasing while keeping a lot of the sharpness spatially. The Baldur line of instruments with all of the above-mentioned qualities makes them perfect for industrial applications.

When selecting a hyperspectral camera for your application, you should define the minimal object you want to detect and choose at least two effective pixels per object (in one direction), then the size of, for example, your conveyor belt, will decide how many effective pixels, and thus, native pixels for that systems, is needed for your application. 

It’s the effective pixel count for an imaging system that is the key quality parameter that is driving the price. Therefore Price/effective pixel is a good measure of pricing for any imaging spectrometer. We believe the Baldur line of cameras to be the most affordable solution per effective pixel on the market today.

No alt text provided for this image

The machine above is a fish quality analyzer using the Baldur V-1024N camera. The machine has been developed in a project called KVASS, which has been funded by FHF. The partners in the project are Maritech, NOFIMA, Lillebakk, Lerøy, Havfisk, Prediktera and HySpex. This is today operational in the Norwegian fish industry. The commercial product is sold by Maritech in Norway.