By Dr. Hans Courrier, Research Scientist and Garrett Grimm, Quality Engineer - March 15, 2024
If a picture is worth a thousand words, what is a hyperspectral image capable
of revealing? As discussed here, a conventional digital
color camera provides 3 optical channels (colors) per pixel. Multispectral
cameras provide more information, with between 4 - 16 channels. A hyperspectral
camera, which typically has over 100 channels, provides a great deal more information. This additional information
can be used to achieve results not possible with 3-color or multispectral cameras.
Even with all
that data, though, a common complaint about hyperspectral images is that the
images do not look great. In many cases that is true, especially compared to
the high-megapixel images we have become accustomed to.
The Pika XC2
was designed to provide far better imaging than other hyperspectral cameras. As
an example, consider the images of calibration targets shown in Figure 1. The
left image is the best image we were able to obtain with one of our leading
competitor’s hyperspectral cameras. The right image was recorded with a Pika
XC2.
Figure 1: The left image was recorded with a competitor’s hyperspectral camera and the right image was recorded with a Resonon Pika XC2 hyperspectral camera.
Even with a
casual glance, one can see that the Pika XC2 image is superior. But can we
quantify how much better it is?
Modulation Transfer Function (MTF)
The generally accepted metric for image quality is the Modulation Transfer Function, or MTF. A detailed discussion of the MTF is beyond the scope of this blog, but details can be found here and in Reference 1. Briefly, any scene captured by an imager can be decomposed into a basis of sine waves – large features are represented by long (low-frequency) sine waves, while small details are made up of short (high-frequency) sine waves. Sharp edges are composed of sine waves of many different frequencies. The MTF is a measure of how well an imaging system can reconstruct the contrast, or amplitude, of a sine wave for a given spatial frequency.
The MTF curve is normalized so that the “zero frequency” sine wave response is 1. Most imaging systems will have a high MTF (near 1) for low spatial frequency sine waves with the MTF tapering off and approaching zero for high spatial frequency sine waves as wavelengths approach the pixel size of the imager. In simple terms, the higher the MTF value for a particular spatial frequency, the better the system is at resolving features of that size in the scene.
Figure 2: Zoomed in sections from the left and right images of Figure 1, above. Note how the lines become more difficult to distinguish from one another moving from right to left, especially in the competitor’s image on the left.
Figure 2 demonstrates this concept. Note how both the competitor’s camera and the Pika XC2 resolve the black and white lines better when the lines are larger and spaced further apart, but the competitor’s camera blurs the lines much more as they become thinner and spaced closer to together, approximating a higher-frequency spatial sine wave.
Looking at an MTF plot for a camera is not an intuitive way for most people to judge whether the camera will produce a “nice” image or not. However, it is easy to compare MTF plots – the larger the MTF value as a function of spatial frequency, the better.
Figure 3: MTF as a function of spatial frequency for the Resonon Pika XC2 hyperspectral camera and a competitor’s hyperspectral camera.
Figure 3 shows the MTF plots for our competition’s hyperspectral camera and for Resonon’s Pika XC2. Note that the MTF of the Pika XC2 is significantly higher than the competitor’s camera as spatial frequency increases. This means that the Pika XC2 will be able to better resolve fine details. The MTF comparison provides a quantitative measure of what is obvious to the eye in the figures above; the Pika XC2 has superior imaging capability.
Does it Matter?
The primary
advantage of hyperspectral data is that the high-resolution spectral
information enables one to distinguish between objects (e.g. healthy and
unhealthy plants or foreign objects in a production line) or states of an
object (e.g. cooked or undercooked cookies) that cannot be accomplished with
conventional 3-color or even multispectral images. With many hyperspectral
cameras, the cost of high-resolution spectral information is low-resolution
spatial information (i.e., poor images).
The Pika
XC2’s excellent imaging allows one to take advantage of high-resolution
spectral information (447 channels), even for very small features in a scene. As
a result, the Pika XC2 is popular with our customers conducting cutting-edge research.
If your project requires detecting subtle spectral differences in small regions,
the Pika XC2 is designed for you.
Our
knowledgeable Sales Team can help you gain
insights worth much more than a thousand words from your hyperspectral data.
Contact us to discuss the specific requirements of your application.
Dr. Hans Courrier, Senior Scientist
Dr. Hans Courrier, Senior Scientist at Resonon
Hans Courrier is a Senior Scientist at Resonon, specializing in remote sensing and innovative applications for hyperspectral imagers.
He holds a Ph.D. in Physics from Montana State University, where his research focused on constructing space hardware for spectral observations of the extreme ultraviolet solar atmosphere and analyzing the resulting data.
Hans’s expertise ranges from radiometric modeling and signal analysis to opto-mechanical tolerancing and design. He has made significant contributions to the ARCSTONE CubeSat project, aimed at improving exo-atmospheric satellite calibration through high-accuracy observations of lunar reflectance, and has authored multiple publications on the topic.
References
1. Modulation Transfer Function in Optical and Electro-Optical Systems
