»
Hyperspectral cameras - HSI¶
FAQ¶
Introduction¶
Q: What is the special technology that differentiates xiSpec from other hyperspectral cameras?
The core technology here is based on applying Fabry-Perot interferometric spectral filters directly on top of every image sensor pixels.
This is a unique approach of adding Hyper spectral filter at wafer-level to integrate virtually multiple spectrophotometers into the format of an imager with multiple spectral pixel arrays that can be arranged in any possible pattern layout, driven by the final application.
The whole process is done by the sensor supplier imec during the wafer production on the micro level and allows incredible miniaturization of HSI camera (Hyperspectral Imaging) making it light and portable.
Hand in hand follows the reduction of the cost thus providing a combination that is making this hyperspectral technology more accessible for a plethora of various application fields.
Q: Is this an "off-the-shelf", commercially available hyperspectral imager solution?
After several years (since 2014) it is a standard product which in 2020 went through an intense upgrade.
The camera is essentially a component that requires integration efforts for utilization in particular applications, but the overall technology of Hyperspectral imaging is known for decades and the USB3 platform is also industry proven and reliable.
Moreover, there is certain standardization that applies and from which the customer can choose, like:
150 bands line-scan spectral imager:
As linescan implies, a translation movement is needed here to capture the hyperspectral image.
So using this camera requires a synchronized camera / object movement and image acquisition.
The different hyperspectral bands are realized in a vertical direction.
After scanning, both a high spatial as well as high spectral resolution is achieved.
With one single scan, you get 150 spectral images between 400 and 1000 nm (3 nm step), each with a size of 2048 x 5 pixels.
Scanning speed is relatively high as it can reach the acquisition rate of 2880 lines/s.
To simulate a regular linescan camera inspection of a small line a special lens/filter has to be used.
10, 15 or 24 bands snapshot mosaic spectral imager:
No translation movement is needed here as this snapshot design enables direct capture of hyperspectral imaging videos for applications such as surveillance or medical in real-time.
This is the last generation of sensor architecture where one spectral filter ‘per-pixel’ is processed on a full mosaic pattern of 4x4 = 10 and 15 channels or 5x5 = 24 channels and repeated continuously on the sensor surface.
No optical duplicator is needed.
HSI data-cubes can be captured at video-rate with a spatial resolution of 512 x 272 or 409 × 217 for each of the 15/24 spectral bands.
The original sensor resolution can be interpolated.
Additional spatial resolution can be reconstructed with specific de-mosaicing algorithms.
Q: Is the camera ready to use out of the box?
For this kind of product, it is critical to be familiar with the hyperspectral technology itself.
The camera delivers RAW data in a form of a datacube, which next needs to be processed by different types of software.
The type of software that the customer should use depends on the application specifics or basically on the materials that need to be analyzed.
So either imec's HSI-Mosaic suite API is required or a separate 3rd party software library could be used.
In certain unique cases, the customer would even have to program their own application software or write some algorithms.
Furthermore, there are certain procedures and accessories that are to be implemented while utilizing Hyperspectral cameras, like a set of high and low pass filters that have to be used to retain second order response of the sensor filters.
For applications that do not require spectral analysis (like classification), the full active range can be used.
Having experience in the field of Hyperspectral imaging definitely helps or is even a condition when working with such a camera.
Clarification¶
Q: Can the range and FWHM (Full width at half maximum) of the 24 spectral bands be defined?
Important to understand is that filters are not attached or "glued" to the sensor - they are deposited ‘monolithically’ at wafer-level onto the CMOS sensor's pixel directly, with the same equipment/process and fab infrastructure which also allows manufacturing of standard CMOS image sensors.
So the design of the filter is fixed - the cavity is fully solid and reliable.
This means the sensors are NOT tunable anymore once processed.
Today there are 4 types of sensors available “off-the-shelf” (please check with our Sales team for details) and more planned to be added to the roadmap during the coming years.
Q: Why do these cameras have a limitation at the spectral range of around 1000nm?
The sensor technology is based on standard CMOS sensors, with a native resolution of 2048 × 1088.
These sensors use silicon as a substrate which becomes transparent at 1100-1200nm and there is no way to detect at 1500 nm for example.
Q: How do you calibrate signal to noise ratio (SNR) and is the calibration required frequently?
A set of documents is provided with the camera (including the Manual) that helps to start with the integration process as well as calibration and other steps.
Please contact info[at]ximea.com for more details.
Q: How does the Hyperspectral imaging work?
Here is a simplified general description:
For every pixel in an image, a hyperspectral camera collects the spectrum of electromagnetic radiation in hundreds of narrow wavelength bands.
As each band can be less than 10 nm wide, the spectrum appears to be continuous.
The range of the spectrum is spread wide from ultraviolet (350nm) to infrared (2500nm) and different hyperspectral techniques and/or sensors operate only on certain parts of it.
For example, Hyperspectral imaging can look at the near-infrared region (750nm and up) which can be far superior to the human eye that can see only visible light in the bands of the three primary colors.
On the output from this type of analysis, you get an image “cube” for each pixel – a 3D array showing which wavelengths of the spectrum are represented, and in what distribution.
These can be connected to the known spectra database of various materials or their states.
That is where the term Hyperspectral signature (of each material) comes from.
Based on this specific signature you are able to distinguish the particular materials or their state (for example the rotten part of the apple from the rest of the apple) - both organic and non-organic.
Further appeal of Hyperspectral imaging is also in its non-invasive nature and real-time results.