Two IR pictures of the same object but with different readings

[morningglory747]

Hello everyone (again),

For those who are aware, I'm the guy that’s trying to take flower temperatures with a thermal expert Q1 for a work-related project . Since I have little to no knowledge about thermal imaging just like everybody with whom I work with, I have but no choice to take more of your time for my troubles. So here it is

https://imgur.com/a/O5dvzvv

I have here two near identical pictures of a bunch of flowers. They are both taken at the same time (like with 10 seconds interval), but what would differ would probably be my exact position and the resolution of the thermal camera relative to the flowers. I think this is where my problem lies. 

I’m using the thermal expert image processing software since it’s the only one I know how to work. I've marked in polygons and a dot different structures that are easily identifiable on the pictures. These markers aren’t exactly the same in both pictures, but they are roughly equivalent and what matters to me is the average (AVG) and not the min nor the max.

What saddens me is that both picture do not show equivalent temperature readings, and by what I consider to be a large amount. I tried to see if I could apply a corrector to one of the pictures, but the ratio varies for every structure and taking a mean of all the data doesn't fully correct it.

This is a bother, because I can’t tell which one of the two is the picture that depicts reality (as close as it can be, of course).

Has anybody ever encountered this problem and what do you think would be the proper way for me to deal with this?

Thank you!

[Vipitis]

Yeah, that sucks. But good to know the solution works for your experiment... Almost.

What my first angle of attack would be is field shutter correction. This is a known phenomenon for the flir Lepton and has been tested a lot in some 400 pages of posts. The general result was that temperature readings after the shutter are up to 3° (10%) too high for 2-5 seconds and stable for the next 45(untill shutter). And operating the camera over a long time(60minutes) increases the sensor temperature and sensor readings by another 5% over time.

I don't know how those principles apply to the Thermal Expert. But here are my ideas: wait 3 or more seconds after you turn it on before taking a capture. Use a reference blackbody and fitting emissivity to calibrate your results and check overtime.
There might be an inaccuracy on the readings within the device specifications. Like it can tell the difference by 30mK but is only accurate to 1.5 °C +/-

Found the reference: https://www.eevblog.com/forum/thermal-imaging/flir-one-thermal-imaging-camera-teardown-and-hacks/msg848878/#msg848878 (may not apply to theremal expert, but could be similar)

[Bill W]

Agree it is a shutter correction problem as the whole image is moved by 2 degrees from the other.

If you need accuracy you really need a reference in the picture, but that could be as simple as a fairly solid metal block / box, painted matt black at around the flower temperature, and you measure the temperature of that with a normal thermocouple meter.

You then measure the parts of the flowers as 'different by x.x°C' with the camera image & software, and correct for the reference temperature.
Matt black = good emissivity and very low reflections
Metal = good conductivity so the surface is uniform and the same as you measure with the thermocouple
Soild / heavy = high thermal mass so it does not change quickly.  Could consider a heatsink block


Bill

[pauledd]

They state:

Temperature measurement error of Thermal Expert is less than ±3℃ or ±3% at ambient temperature.
http://www.i3-thermalexpert.com/support/faq_technical/

[Pinkus]

also to take in mind: if you moved close to the flower to make the photos, the IR photo will/might include reflected heat (from your body). And as you move, it is not always the same.
Everyone knows IR reflections on glass (e.g. windows), but reflections occur almost on all materials (only not so visible).
To exclude this possibility, i would recommend to put the camera on a tripod and to release it via automatic timer / remote release and try it again.
Also the exact temperature is not what IR cameras will show, they are exact with delta T, not with absolute T.

[morningglory747]

Hm, interesting. I wasn't aware of these issues even though I spent quite a while looking up at how to take pictures with IR cameras. I have heard of the importance of needing a reference temperature, but since I was operating alone, I had hoped that the absence of a reference temperature wouldn't of been that critical. Guess I was wrong.

I was aware of the 3% error and most of my pictures were taken at around 20-25 degrees celcius (68-77F), so atleast it's not as bad as a 3degree error, but perhaps the incident solar radiation would also accentuate the error?

Indeed knowing the air temperature is also quite important! Luckily there is a meteorological station near the orchard that was taking air temperature readings during my measurements.

And operating the camera over a long time(60minutes) increases the sensor temperature and sensor readings by another 5% over time.

.
The thermal expert app would frequently crash (once or twice every 5-10 minutes, sometimes even more frequently), but I was able to use for up to 2 hours before running out of cell battery. Do you believe that this effect would still apply?

Thank you all for your knowledge

[morningglory747]

That is true, but since I was outside, I was hoping that the turbulence of the air would dissipate the effect. I did try using a sort of improvised tripod made out of wood because I didnt have acces to a real one during the time where the flowers were present, but the flowers would always be of different height and angle so it was time consumming for me to always re-adjust the camera on the tripod in order to get the best shots.

[morningglory747]

So I checked a bit on shutter correction and, from what I understand, it seems like thermal expert does have something to take this into consideration. Basically, every time the software gets opened on the android, a calibration is required where I’ll use a lens caps to cover the len while the calibration occurs. This would usually take 10-15 seconds and would produce a much clearer image with reduced noise.

I'm not sure if that counts as a successful shutter correction or not though.

Thanks!

[Tanya]

Hello,
I have a question regarding the reference object, so when taking the image with a reference object we need to  adjust for emissivity difference which will be done afterwards in the post processing, right?. I usually use a matt black sheet of paper as a reference but I am having a problem reading its temperature using a thermocouple, I tried a polymide tape to secure the tip to read off the temperature from sheet surface but I am not having the same temperature after adjusting the emissivity  .

[mahony]

I do not know how long the camera was running when the images where taken. Maybe directly after boot? This may cause such huge temperature difference in very short time.
From my experience with the TE-Q1 is that after turn on you should ultimately run the camera for roughly 2, even better 3 minutes or longer so that everything internally is at or close to a steady state. The same is true if you have significant change in ambient conditions (i.e. moving from indoors 18°C room temp to 35°C outdoors). Especially directly after boot there is a significant temperature gradient in FPA temperature and also for the image temperature readings.

As for the reference: I just build two simple calibrators using a aluminium and/or copper block (40x40x10mm and 50x50x10mm) painted flat black with 3 holes roughly 10mm deep that hold 3 digital thermometers (DS18B20) which I read through an Arduino/Teensy microcontroller. To get a better estimate of current temperature I do averaging of all sensors and some low pass filtering of the data. For my purpose I am regulating the temperature by a peltier element attached to the block and some large CPU heatsink/cooler. For you it should be fine to just measure the temperature I think. This could be package quite small together with some LCD or hooked up to tablet to display the current temperature reading(s).

For in-field use I would probably shade the setup to avoid sudden solar induced heating of the calibrator block.

Edit: Attached is a plot of FPA temperature (orange) and temperature readings from the image center of a 'const' source (room wall) from boot up to ~520 seconds just to get an idea of the amount of temperature drift until stabilization.

[morningglory747]

Sorry for the late reply and thanks for your great answers Mahony.

Darn if only it'd known sooner about this! I honestly didnt expect the adjustement to be this severe. I sure feel silly now. Altough it would of been nice for the TE manual to mention this obviously necessary period of ajustement before being able to safely take shots.

The setup you're proposing looks pretty legit and cheap. If I still have my job by the next flower season of next year, i'll make sure to try out something similar!

Thank you all, 

[mahony]

Your welcome!
Also took some time for me to see this thread and reply as I usually do not check the forum on a daily basis.
If you need something for the calibrator I can supply you with my Arduino sketch and images if required. Just ask here of via PN (might be better as I get an email notification then).

[Bill W]

Hello,
I have a question regarding the reference object, so when taking the image with a reference object we need to  adjust for emissivity difference which will be done afterwards in the post processing, right?. I usually use a matt black sheet of paper as a reference but I am having a problem reading its temperature using a thermocouple, I tried a polymide tape to secure the tip to read off the temperature from sheet surface but I am not having the same temperature after adjusting the emissivity  .

The reason for using matt black is so the emissivity is 1, and then you don't need to know about anything else.  Even 0.99 is close enough as long as the reflection is not the sun :-)
A metal block allows the temperature probe to be nicely buried and so the metal will dominate any other effects on the probe temperature.  You might have conduction down the leads for example and on just paper that might be significant

Bill