Hyperspectral imaging combines both imaging and spectroscopy. There are many different optical architectures that are used to make hyperspectral systems, but the end goal is the same – to make an image where each pixel in the image contains information from many spectral bands (many different colors or wavelengths).

Making good hyperspectral systems is not easy, but, if you only have quality in mind, your highest priority should be the spectral fidelity in each pixel – i.e. the spectrum captured by one pixel is an actual physical representation of the scene imaged by that specific pixel. There are many key quality parameters that influence the spectral fidelity of a hyperspectral system. The influence on the spectral fidelity from the different key-quality parameters depends on many factors but, in general, the most important parameters are:

  • Spatial misregistration

  • Spectral misregistration

  • Straylight

  • F#

  • Spectral response function

  • Detector parameters

  • Environmental stability

To evaluate the performance of a hyperspectral system, we need to introduce the term point-spread function (PSF). This is the shape of the intensity curve for the energy that hits the detector through the optics from a point source.

Hyperspectral camera calibration.

Ideally, in a pushbroom hyperspectral system, the spatial sampling is the same for all bands. In reality, this is never true and results in spatial misregistration.

For one position in the FOV, the shape, size, and position of the center of gravity for the PSF should be the same across all bands. A design goal is to make the PSF shape and size as similar as possible for all positions in the FOV. In real-life applications, the spatial misregistration itself1 and the different sources causing spatial misregistration2 are of great importance, and it can be shown how differences in its gravity3 can compromise data processing results.

Maximum errors in a scene caused by keystone (red), by differences in PSF width (blue), and by differences in PSF shape (green) for 15 simulated cameras.2

Spectral misregistration is also very important for the spectral fidelity of a hyperspectral system. As with spatial misregistration, the shape, size, and position of the spectral PSF is important. To avoid spectral aliasing, the width of the PSF should be kept close to two spectral bands, but in many situations, sharper optics than this are preferred.

There are many reasons for making sharp optics per pixel and per band. For any given detector, with a given detector pitch, the imaging spectrometer will always output more information with sharp optics. On the other hand, sampling the PSF with more than one pixel gives more information about the PSF, but for a given detector, it will reduce the resolution of the system. Any given optical system will always output more information, the more pixels you have per point spread function (PSF)4.

Hyperspectral camera integrated with a LiDAR solution.

There are many trade-offs when deciding what system to make – or for the user, what systems to buy. The sharpness of the optics is one of them.

Another key-quality parameter is the straylight of the optical system. In general, straylight is light that reaches the detector from places where it should not. This can be due to reflections inside the optics, scattering from optical surfaces, reflections on mechanical surfaces, and so on. The straylight effect can also be different for different spectral ranges. There is no industry standard on how to measure the straylight of hyperspectral systems today and different suppliers do it differently.

The light gathering capability of the optics is usually a very important parameter (low F#). The more spatial pixels and spectral bands in the system, the less light that reaches each pixel/band. This means that a high-resolution hyperspectral imaging system is required to be very light sensitive. This parameter can vary in importance depending on the platform. If the hyperspectral system is deployed on an aircraft or in an industrial setting, speed is usually an important parameter and a low F# is then very important for getting a good signal-to-noise ratio (SNR) for the whole spectral range. Conversely, lab applications can typically collect more light by using slower image capture rates and by integrating longer, which allows for more flexibility on the aperture.

Final radiometric calibration at the end of FAT. It is important that the calibration is traceable and stable.

High SNR for the whole spectral range is a very important parameter of the hyperspectral system and how high SNR is depends on the spectral response function of the whole system. The peak SNR only specifies the maximum SNR from a band that is close to saturation and, therefore, does not give the whole story. One would also need to know what the total quantum efficiency of the whole system is as a function of wavelength. To provide useful information, the SNR curve needs to be specified for a given input radiance and a given (and operationally realistic/relevant) integration time/exposure.

The detector is an important part of hyperspectral systems and, in many cases, defines the design goal of the optical systems. The SWIR spectral range (900-2500nm) has a very limited number of available detectors and the price increases drastically with increasing pixel count of the detector. This is a scenario where the optics are designed for the detector and usually designed to be as sharp as possible.

For the VNIR range (400-1000nm) there are a lot of detectors available and the detector which makes the most sense for the optical system can be chosen. For both the VNIR and SWIR range, there are many parameters of the detector that are of great importance. These are the full well capacity, noise floor, read-out modes, read-out speed, pixel pitch, quantum efficiency as a function of wavelength, and many more. The detector market is constantly changing, and it is important for hyperspectral manufacturers to keep up with the detector market to integrate the best detectors available.

Hyperspectral imager mounted on a UAV in Cuprite Hills, Nevada. (Photo courtesy of Echo Labs, Canada.)

Calibration procedures and standards used (including accuracies) should be available for the users and having radiometric calibration traceable to NIST or PTB standards (or similar) is, therefore, very important.

Any hyperspectral system needs to maintain a stable and accurate radiometric and spectral calibration outside of a controlled environment. It is worthless having a perfectly calibrated system leaving the factory if it is not stable and valid after transport and during operations. This means the spectral, radiometric, and geometric calibration must be stable within different temperatures, pressures, and under heavy vibrations, in order to ensure the system will provide repeatable and reliable results under demanding conditions, such as UAV operations.5,6

A mosaic of 5 hyperspectral flight lines.

It is very difficult to discriminate different hyperspectral systems from suppliers’ top-level datasheets. When comparing systems from different manufacturers (or different models from the same manufacturer), a detailed report specifying the aforementioned parameters for that particular camera model should be provided by the supplier. Additionally, it is advisable to request sample data from a scene relevant for the user’s application.

There is currently an action to make a common standard for characterizing hyperspectral cameras, organized by IEEE7. HySpex is supporting and heavily involved in this group, aiming at making the offerings more transparent for end users of hyperspectral imaging systems.

This article was written by Trond Løke, CEO, Norsk Elektro Optikk, AS (Oslo, Norway). For more information, contact This email address is being protected from spambots. You need JavaScript enabled to view it., or visit here  .

References

  1. https://www.spiedigitallibrary.org/conference-proceedings-of-spie/8706/1/Resampling-in-hyperspectral-cameras-as-an-alternative-to-correcting-keystone/10.1117/12.2015491.short
  2. https://www.spiedigitallibrary.org/journals/Optical-Engineering/volume-59/issue-08/084103/Spatial-misregistration-in-hyperspectral-cameras-lab-characterization-and-impact/10.1117/1.OE.59.8.084103.full?SSO=1
  3. https://www.hyspex.com/keystone/
  4. https://www.hyspex.com/sharp_optics_many_pixels/
  5. https://www.hyspex.com/scientific_grade_uav/
  6. https://www.hyspex.com/quality_vs620/
  7. https://standards.ieee.org/project/4001.html