Hyperspectral Imaging

[colinbm]

Hyperspectral imaging is here....

https://cubert-gmbh.com/

[Andy Perrin]

There are lots of hyperspectral cameras out there at a range of prices (the range goes from “high” to “astronomical”). I would love to get one that works outside of visible.

[colinbm]

This one says 350 - 1000nm.

[Andrea B.]

Is there a definition somewhere? Like, hyperspectral vs multispectral?

[Stefano]

Here's an explanation: https://www.google.com/url?sa=t&source=web&rct=j&url=https://gisgeography.com/multispectral-vs-hyperspectral-imagery-explained/&ved=2ahUKEwiY3-6WpLbyAhXZhf0HHU6qA7QQFnoECCwQAQ&usg=AOvVaw2UdgCT05uuytA5wJeJp16e

[Andrea B.]

Are the bands recorded all at once? Can you then extract a UV photo or an IR photo from the data?

[Andy Perrin]

Are the bands recorded all at once? Can you then extract a UV photo or an IR photo from the data?

Yes exactly! Indeed you can extract a UV photo or IR photo, with the appropriate algorithms. Or any other slice you want. Or any three slices for a tricolor. You can filter your images AFTER taking the photo, in software...

 

Another name for a hyperspectral camera is an imaging spectrometer. (It depends on whether you want to think of the device as a camera that takes snapshots at different wavelengths, or as a spectrometer with pixels.)

[Stefano]

It's like having a spectrometer for each pixel. Such a camera would make all sorts of aerochrome or odd image processing much easier. Having one that can see between 300 and 400 nm would make it possible to see glass slowly darkening in deeper UV, for example. I don't think I will ever afford one, thought.

[Andy Perrin]

As you'd imagine, the spatial resolution tends to be very low for these cameras.

[Stefano]

That's the biggest downside, together with the price. You would need to make the single subpixels very small and have a very high raw resolution. We got to the point of having 100 megapixels on a phone, so probably we can make a larger sensor with a very high pixel density, but it's not cheap. In general, a bigger sensor, as well as a bigger CPU has a higher probability of having defects.

[Andrea B.]

Indeed you can extract a UV photo or IR photo, with the appropriate algorithms. Or any other slice you want. Or any three slices for a tricolor. You can filter your images AFTER taking the photo, in software.

 

This would be very fun and cool.

 

 

I think I see the gear better as an imaging spectrometer.

 

I wonder how quickly the data can be gathered? Maybe the imaging area is kept small because it takes a while to run through 50 wavelength settings?

 

And how much does such a "camera" cost anyway?

[Andy Perrin]

I wonder how quickly the data can be gathered? Maybe the imaging area is kept small because it takes a while to run through 50 wavelength settings?

 

And how much does such a "camera" cost anyway?

 

There are many many varieties of ways to build hyperspectral cameras. Some of them use variable wavelength LCD filters to run through 50 wavelengths or whatever, but others use a diffraction grating and capture 400 line spectrums and then combine them into an image! Others use 2D diffraction gratings, and there are loads more variations. Nobody has settled on a unique way to build one.

 

MaxMax has one, which uses a linear variable bandpass filter to filter the light from a single image in stripes, and then builds up a panorama by rotating the camera IIRC. Just $14,000!

https://maxmax.com/m...e/hyperspectral

[Andrea B.]

I think perhaps I'm not completely impressed by the MM multispectral results. But OTOH, MM has written the software and so forth, so it is quite interesting.

 

slightly off topic: I can't remember what this is called --> that to create a "proper" pano you must rotate around a certain vertical axis of the camera.

[Daniel Csati]

Here is a example of a hyperspectral image, scanning the wavelengths from 500-930nm. It was a pushbroom type imager (I sold it already). Heavy and not easy to use. Spatial resolution is limited to a MP or two but if you have the time and 3kUSD to build it, it's fun.

It can be optimized for UV too..

 

[Andy Perrin]

Very cool, you can see the Wood effect kick in. 

[OlDoinyo]

I wonder what the wavelength resolution is. If it is around 1 nanometer, the file size would be equal to that of 650 normal monochrome frames--some hefty image files there, especially if uncompressed. 

 

It would be the digital equivalent of a Lippmann photograph--color rendition without assuming anything about the viewer's color vision biology.

[Daniel Csati]

The raw data was something like 30-50GB pro data cube. A final, processed data cube at 1080p resolution and 160 layers was about 1.2-1.5GB. The easiest format to use is a multilayer .tiff, which can be even further processed in Scyven.. Recognition of materials, identification of source light spectra, calculation of reflectances. A lot of data to mine valuable information..

 

The real wavelength resolution was only 2-3nm but I sampled it 10-15x more than needed :)

[Doppler9000]

On 8/17/2021 at 3:20 PM, Andrea B. said:

slightly off topic: I can't remember what this is called --> that to create a "proper" pano you must rotate around a certain vertical axis of the camera.

It seems you were referring to the entrance pupil of the lens.