LWIR, MWIR & SWIR side by side

[Ultrapurple]

I had a little time to call my own today and some new-ish toys to play with so I thought I'd try comparing imaging at three different wavelengths: LWIR (7-14µm), MWIR (3-5µm) and SWIR (0.9-1.9µm).

I used a Therm-App Pro (640 x 480) with 19mm lens for LWIR, an Agema THV 450 (120 x 120) with 20 degree lens for MWIR and an ElectroPhysics  MicronViewer 7920A (see later for resolution).

There is a tremendous size difference between the three cameras. The oldest, the mid-1980s Agema, is the size of an early ENG camera-recorder. It represented a huge technological advance, for its time, being one of the very first completely portable MWIR thermal cameras that could operate from batteries. Unlike many cameras of its time it didn't require liquid nitrogen or a limited-life Stirling cooler; instead, it used a single-pixel 'SPRITE' detector (cooled to about -70°C by a stack of Peltier devices), with the raster image being got via a system of moving mirrors. There is plenty of documentation on the web (particularly on EEVblog, thanks to Fraser, who did a very well-illustrated and detailed teardown of the very-similar THV 470). It produces an image of 120 lines x about 120 pixels at 20Hz, 'dropped in' to a standard PAL TV frame (along with image data) by an integral 7-bit scan converter circuit. I'm still learning to drive this camera and it may be possible to get better-quality results than in these first experiments.

In the middle, size-wise, is the ElectroPhysics MicronViewer 7920A. These are still relatively current, having only gone out of production in the last few years as the supply of a critical part dried up. Despite that, the technology really has its roots back in the 1960s, possibly earlier, as the camera is based around a vidicon tube. Unlike the pyroelectric vidicons in first-generation fire cameras (waves at Argus 1), the short-wave (red to about 1.9µm) tube doesn't require a chopper or a fancy frame store; it's basically the same technology as you'd find in a Pye Lynx. Vidicons are normally operated at ambient temperature - if you feel comfy, the tube will too. This camera uses the NTSC (525/60Hz) video standard which means it's a bit difficult to express the resolution in good ol' computer-friendly pixels. It is, roughly speaking, 640 x 480 but until I delve into the tube data sheet I won't be able to say exactly what the true optical resolution is. But 640 x 480 is a good-enough place to start.

The LWIR camera is the newest, and smallest of the gang: a Therm-App Pro, 640 x 480 25Hz Android dongle-style camera.

I used a cheap USB video capture dongle running on a phone to digitise the MWIR and SWIR video outputs. I adjusted the black level, gain and (particularly) gamma of the resulting images to give what I felt was the best possible result.

Let's start with a look at the three cameras side by side.

[Ultrapurple]

So, what about the images? Here is my standard test target, my little shed. You can see what I mean about the MWIR resolution being well behind the others - but the camera (and its technology) is nearly 35 years old.

[Ultrapurple]

Now a pile of trash in front of a near-derelict conservatory that's full of more stuff that I need to get rid of. Interestingly, the glass is transparent to the SWIR camera. I think it was opaque to the MWIR camera but the resolution wasn't really good enough to be sure.

[Ultrapurple]

And last in this series, a selfie. Notice my spectacle lenses are opaque at LWIR and MWIR but transparent to SWIR.

[Ultrapurple]

Finally, here are three unretouched images exactly as they came from the respective cameras.

The three cameras have different fields of view so it isn't really possible to do a 'perfect' side by side comparison. I compromised by having the cameras the same distance from the 'still life' subjects but adjusted my selfie position till it looked 'about right'. I'm not convinced I got the focus spot-on for any of the selfies except the LWIR one.

And that, my friends, was an afternoon's work. I had hoped to increase the spectral range to include near-IR (0.75-1.1µm), visible, plus UV (380-320nm) but time was against me. Another day, perhaps.

[Ultrapurple]

Just a further note on the SWIR camera. It is presently fitted with a good quality C mount lens with manual focus and aperture. I have placed a 950nm longpass filter over the lens to ensure that its images aren't too contaminated by near-IR.

I experimented with the aperture and, as expected, shut down to f/16 there's very little to see, even in bright sunshine. I kept the gain control set to automatic and opened the aperture stop by stop while watching the picture. As more light was admitted the picture 'came in' from very low contrast to decent quality by about f/8 or f/5.6. Opening the aperture further didn't appear to improve the image so I kept it stopped down to the smallest aperture commensurate with a good image. I am always wary of harming the pickup tube by exposure to too-bright light. On tube cameras in particular I normally keep the aperture stopped down to a minimum when not in use and also always use a lens cap. I believe vidicon-type cameras can suffer from image burn-in even if they're not switched on, so its a sensible precaution.

I didn't find the camera particularly laggy, certainly no more so than 'most' vidicons. Generally speaking, vidicons get laggier as the illumination level falls, so using this one with the 'right' amount of light certainly helped.

[Vipitis]

Uh, I have to read this again in depth but I love the comparisons. To me MWIR and LWIR had a certain look, in terms of how reflective it feels and how humans look.

As I said in the other thread, I want to try the chromatic order stuff (SWIR=Blue, MWIR=Green and LWIR =Red) to get a 'color image'. I know that two color MWIR/LWIR devices exist and were tested for military use but I am not aware of any 3 color build. There will be loads of issues due to parallax and other differences but the first attempts at color photography were equally crazy. I have gotten the cables I mentioned and spent the last hour today to figure out how to get the full spectrum camera to operate and got something from it. Next will be the LWIR cores bug it's quite promising.

Perhaps you got video samples? Would love to try my hand at compositing them.

[DaJMasta]

Nice work.  My efforts to do something similar have stagnated around trying to better communicate with my MWIR camera, but I'm hoping to get back to it in the next couple of months.

My camera setup is basically tethered to my computer, so it's harder to get at outdoor targets, and I use much lower light levels for SWIR and UV, but it's fascinating to look at things in different bands.  I wonder if there are household materials that are transparent in LWIR or MWIR but not the other - it's always fun to see the comparison side by side, and it seems like that's often the best way to demonstrate.

[CatalinaWOW]

Unless you had some very special glass in your conservatory it was opaque in the MWIR.  Most common glasses cut off around 3 micron. 

For those looking for household materials that show differential transparency in the LWIR and MWIR your best bets are the various thin flexible plastics such as are used in many bottles and the like.  The actual transmission is somewhat sensitive to exact composition and sometimes to processing so it is hard to predict, but it is relatively easy to try.  There are examples that exhibit all of the possible options, relatively transparent in one, the other, both or neither.

"Cameras" have been built which support the type of three color images you describe, though at low frame rate.  The ones I know about are called imaging spectrometers.  You can generate images in any set of "colors" you want by integrating the outputs over the appropriate bands.  Obviously post processing and in all but the rarest cases would be generating single frames, not movies.  The highest resolution one I heard of was 256 x 256, but it wasn't my field and I haven't looked for a long time.  Obviously very expensive toys, and I really only was aware of a handful being built.  All research instruments, not series production items.

[coppercone2]

I think to highlight the differences between the cameras you want to setup an experiment with a smoke screen and try to photograph stuff behind the smoke. thats where I always see the interesting bits about the IR range show up, military smoke screen.

making a few different kinds of smoke (organic residues (black), metal oxides (burning metal), boiled wax, fog machine) and maybe other obstructions (throwing a handful of powdered sugar or flour). I think the main interest in multispectral imaging is target acquisition.

and maybe things like seeing through tissue paper, napkins, etc

[CatalinaWOW]

I think to highlight the differences between the cameras you want to setup an experiment with a smoke screen and try to photograph stuff behind the smoke. thats where I always see the interesting bits about the IR range show up, military smoke screen.

making a few different kinds of smoke (organic residues (black), metal oxides (burning metal), boiled wax, fog machine) and maybe other obstructions (throwing a handful of powdered sugar or flour). I think the main interest in multispectral imaging is target acquisition.

and maybe things like seeing through tissue paper, napkins, etc

Actually, while there has been military interest in multi-spectral imaging the hottest application for it is in crop classification, crop health monitoring, soil type classification and similar things.   Using narrow band filters around the chlorophyll line, the CO2 lines, the water lines and the like.  Images taken from drones, airplanes and satellites.  Lots of literature available.

There are a lot of reasons the military side hasn't gone hog wild on this, and include cost of sensors, worth of the incremental improvement from multi-spectral.  There is benefit, but in most cases it isn't huge and the costs of implementation are huge, both initial investment and the logistics train that follows. 

I believe there is quite a large unclassified literature on the effects of smoke on these various sensors.  The standard references in the field certainly discuss it.  You can get a lot of insight by looking into the basic physics.  DeBye scattering and the mechanisms for molecular absorbtion.

[coppercone2]

looking at plants and weed growths and woods would be interesting too

speaking of smoke the fire department loves these things, the problem is actually not seeing the fire (I heard stories of people crawling into rooms that were totally on fire because they were glued to the TI display). not sure what different smoke conditions cause though.

[Ultrapurple]

Perhaps you got video samples? Would love to try my hand at compositing them.

PM sent.

[Ultrapurple]

setup an experiment with a smoke screen and try to photograph stuff behind the smoke. thats where I always see the interesting bits about the IR range show up, military smoke screen.
making a few different kinds of smoke (organic residues (black), metal oxides (burning metal), boiled wax, fog machine).

Unfortunately my location is not conducive to making smoke of any kind. Even a small, low-smoke fire in a proper brazier upsets the neighbours; anything larger soon get illuminated by flashing blue and red lights.

The image below was made with a near-IR capable camera under (probably) 750nm flood lighting - a standard security camera setup.

[Ultrapurple]

Just for fun I tried reprocessing a 10-second recording of the MWIR image via Autostakkert! to see if I could increase the quality of the image. Judge for yourself - below is an original frame (one of >100) selected at random, plus the reprocessed result.

[coppercone2]

setup an experiment with a smoke screen and try to photograph stuff behind the smoke. thats where I always see the interesting bits about the IR range show up, military smoke screen.
making a few different kinds of smoke (organic residues (black), metal oxides (burning metal), boiled wax, fog machine).

Unfortunately my location is not conducive to making smoke of any kind. Even a small, low-smoke fire in a proper brazier upsets the neighbours; anything larger soon get illuminated by flashing blue and red lights.

The image below was made with a near-IR capable camera under (probably) 750nm flood lighting - a standard security camera setup.

yeah unfortunatley its easy to get in trouble trying to do smoke science even though any developments would be useful as hell

[DaJMasta]

Just for fun I tried reprocessing a 10-second recording of the MWIR image via Autostakkert! to see if I could increase the quality of the image. Judge for yourself - below is an original frame (one of >100) selected at random, plus the reprocessed result.

I don't see a big difference, but I think that's a limitation of the input data - it's unfortunate that the Thermovision 450/470 cameras only read 8 bits from the detector, you could probably get a good bit more image quality from post processing with a few more bits of signal.  Given the age, size, and simpler TEC, though, I guess it's asking a lot 

[Ultrapurple]

I think that's a limitation of the input data - it's unfortunate that the Thermovision 450/470 cameras only read 8 bits from the detector

I must check the manual again but I thought the digitisation was to a lot more than 8 bits - I think I remember 13 or 14 bits. But the readout to the screen is definitely only 7 bits (you can select fewer!)

It's still very early days for me with this camera. I have a feeling something may not be quite right with its scanning setup, as the diagonals appear to be suffering from line tearing (exactly like you get on poor interlaced > non-interlaced video conversion). The earlier SPRITE cameras used 4:1 interlace and I suspect this one is the same, but reading into a framebuffer so you don't normally notice. But if it is writing out of sequence, you get the line tearing we see here.

Investigations will continue...

[Ben321]

Very impressed with your pics. Great SWIR and LWIR pics. LWIR looks to be a VERY high-res camera, like a FLIR T1020 or something. How did you get that? Did you use it in a job you worked at, and they let you take it home? The fact you were able to get your hands on an MWIR camera is also very impressive. MWIR cameras can't really be found cheap anywhere. Cheap SWIR cameras can be found, if they are the vacuum tube based cameras like the Electrophysics Micronviewer, however MWIR cameras I don't think exist in a vacuum tube type. They are all solid state, which like solid state SWIR cameras, are VERY expensive. Sadly, your MWIR camera doesn't have a very high resoution. It appears to be well below 320x240.

[Ultrapurple]

Thanks Ben321. The LWIR camera is 640×480 native with resolution enhancement to 1280x960. Good results, though, also stem from practice.

The MWIR camera is only 120x120 or so. As you say, MWIR can be extremely expensive - and, as has been discussed in many other threads, normally requires cryogenic cooling by a limited life Stirling device. The advantage of the single pixel SPRITE based sensor is that it is cooled by stacked Peltier devices, an approach that is traditionally not practical for staring arrays.

[bap2703]

Interestingly, the glass is transparent to the SWIR camera.

You should try your visible photo lenses on that one.
I tried a couple Canon EF lenses on a swir camera, and I found that my 100mm macro works fine (tested with a laser at 1550 nm).
Some others don't and give a blurry image with a lot of "diffused" light.

[Ultrapurple]

I am pretty sure some of my visible lenses will work just fine with the SWIR camera (but see 'hot spot' caveat later), however I think focal length will be an issue. A typical CCTV lens for a 1" tube has a focal length of perhaps 10 to 20mm for a 'normal' field of view. Although I do have 35mm camera lenses down to 8mm (the famous f/2.8 fisheye, which is great for one-shot 180 degree pictures that you can unwrap into an all-round panorama), most are in the 50-200mm range - which would be extreme telephoto for the SWIR camera.

Adapting from the C mount camera to a Nikon F mount lens wouldn't be too difficult, as the F mount back focus distance comfortably exceeds that of C mount. Somewhere I've got an adapter that goes the other way round, to put C mount lenses on a Nikon body. It doesn't work well, as 'infinity' is something like six inches (in most cases).

Many perfectly good visible light lens designs suffer from a 'hot spot' (or 'bloom') in the middle when used at near-IR, such as this minor example:



I soon switched to using lenses that didn't suffer from this effect so I don't have all that many example images. But I do think it's likely to be just as much a problem at SWIR - and possibly more so.

[coppercone2]

michael crichton's congo I think had alot of stuff about thermal imaging trees from elevations to determine something about stage of growth or logging activity. it might be interesting to take pictures of a tree canopy from high up to see if it shows you anything.. his books are usually decent, but I forgot if it was about AI or thermal imagining but it sounds relevant

[Ultrapurple]

It's a long time since I read Congo but I do recall being somewhat fazed by the description of a top-notch computer (functionally speaking it was a PC- or laptop-equivalent) that had some (then-impressive) amount of RAM that couldn't be rationally explained by anyone conversant with powers-of-two. Something like 82KB rings a distant bell - 64K + 16K + 2K. Why???

Now of course someone will go read the book and tell me that it was because that particular chipset worked in base 3, or something.


I don't recall it mentioning thermal imaging but I have no reason to think you're mistaken. I read the book before I developed an interest in thermal imaging - and well before I realised that used thermal imaging cameras were available at affordable prices.

[Vipitis]

I wanted to post this earlier, but here are the results when combining S,M,L to RGB in chromatic order. I just aligned them by hand as best as possible and tuned contrast for each channel a bit. There are better ways to do the mixing and this was just from a quick test, I most likely won't be able to do the more sophisticated video one any time soon. But I wanted to share this at least.

one observation: shadows are blue because they are "cold" like the emit more in the SWIR compared to the LWIR.

the spectrometer is an important comment and hyperspectral images (data cubes) are really interesting in how you present them. But line scanning infrared spectrometers seem not to be a very available product. However I recently read of an interesting research result of a "tunable pixel" (https://www.nature.com/articles/s41566-021-00787-x) which sounds interesting. Perhaps I should make a thread about current research publications, there was something else thin film that could bring 1.6µm to the visible (http://dx.doi.org/10.1117/1.AP.3.3.036002)