Latest Seek Compact (non-pro) has significantly reduced noise

[IwuzBornanerd]

Either Seek has substantially reduced the noise in their cameras or there is a lot of variation and I got lucky...twice.  After determining that the original XR I bought was useful for my surveillance purposes I bought one with the 36 degree field of view, since I figured that would be suited to one of my scenarios.  It was immediately apparent that this new camera had less noise than the one I bought in 2015--so much so that I had to revise my software to take advantage of it.  The new camera has certainly one quarter as much noise as the old one, if not one eighth the noise!  Images are as clean at 0.2 degree Fahrenheit resolution as they were before at 1 degree resolution, and even 0.1*F is usable, particularly with a smoother palette than my original palette.  I was so happy with the non-XR that I bought a new XR and it's performance is as good at the other new one.

On the new cameras frameID10 == frameID4 and the values are 4 digit values whereas in the old camera the values in frameID10 were 2 and 3 digit values.  With adjustments to my code I can still use frameID10 for correction factors but, as before, the results are not as good as my own "user generated" correction factors.  Another noteworthy difference between old and new is that the curve in frameID9 has been extended beyond the old discontinuity.  This suggests to me that maybe they moved the operating point on the thermistor curve in such a way as to reduce noise susceptibility.

Another factor with the low noise version is that NOW I can see the effect of my bias correction.  I had tried before to correct for the change in pixel bias that occurs far too often but I could not see any effect and I wondered if it was because there was so much noise in the correction factors themselves.  I can see it now and it is very noticeable when the camera temperature is only a few degrees different from what it was when the main correction factors were created.

I uploaded some comparison images.  In each snapshot the old XR I bought in 2015 is on the left and the new XR is on the right.  These were all taken back when it was still chilly outside so the heat was on & it was cold on the other side of the wall

The first image is of my wall corner showing part of the cathedral ceiling & heat register at 2 colors per degree C.

The next image is the same scene at 10 colors per degree C.

The 3rd image is the same but with 20 colors/degree C.  You can see the increasing noise.

The 4th image is 20 colors/C but with my surveillance palette, which is smoother than my standard one, so the image does not look as noisy as the 3rd one.

The palette I use is one I made such that I can clearly see the difference between adjacent colors in the palette, so whatever fraction of a degree I apply to each color will (hopefully) be noticeable in the image.  I have found that this is not true of the palettes used by other software authors.  I loaded frenky's "rain" & "iron" palettes into my palette editing program & deleted colors until I could distinguish each one & that reduced those palettes to 30-50 colors from 1000!

For those of you accustomed to the smoother palettes, the 5th image is 20 colors/C with one of Frenky's palettes.  The smooth palette hides a lot of noise but also loses some detail when there is not a lot of noise.

The 6th image is my mantle wall with my standard palette at 2 colors per degree C.

The 7th image is the same scene at 10 colors per C.  You can clearly distinguish the candlesticks and picture frames, etc. on the mantle.  You may not know what they are, but you see objects there.

The 8th image is the 7th with the bias correction off, so you can see the effect of that.  I admit that the old XR looks better with it off, but it looks bad anyway at that resolution.

The 9th image is veins in my forearm with my standard palette at 10 colors per degree C.

The new cameras are cleaner but there is still room for improvement.  One thing that has room for improvement is that the thing will frequently jump in temperature readings by 1, 2, & even 5 degrees.  This appears to be the entire frame at one time unlike the usual bias change effects, but it does occur with the shutter frame.  It is worse when the camera is operating at below-freezing temperatures and under those conditions it causes false triggers in my surveillance code.  Both frameID1 and frameID3 are changing significantly in those instances but pixel 1 is only 1 different.

[Fraser]

That is a dramatic improvement !

The images from your 2015 model camera are what I am used to seeing and why I sold my SEEK soon after purchase.

The images from your latest purchase are more than acceptable fo many tasks. I would have expected SEEK to make some announcement about releasing a new 'lower noise'' version of their microbolometer. I have not seen anything stating such.

It is possible that SEEK and their microbolometer fabrication house have been making progress in taming the noise in their sensor. When first released, SEEK claimed that they would be developing and releasing a new microbolometer that did not need the Patent pixels to comply with FLIR patent claim. Maybe you are seeing a new microbolometer in your new camera ?

It would be interesting to hear from other owners of the SEEK who purchased recently. Let us hope SEEK have improved their product. I welcome their competition in the market and efforts to make thermal imaging affordable technology for all.

Fraser

[paradigmic]

Either Seek has substantially reduced the noise in their cameras or there is a lot of variation and I got lucky...twice.  After determining that the original XR I bought was useful for my surveillance purposes I bought one with the 36 degree field of view, since I figured that would be suited to one of my scenarios.  It was immediately apparent that this new camera had less noise than the one I bought in 2015--so much so that I had to revise my software to take advantage of it.  The new camera has certainly one quarter as much noise as the old one, if not one eighth the noise!  Images are as clean at 0.2 degree Fahrenheit resolution as they were before at 1 degree resolution, and even 0.1*F is usable, particularly with a smoother palette than my original palette.  I was so happy with the non-XR that I bought a new XR and it's performance is as good at the other new one.

Have you connected your original 2015 Seek to a phone running the latest Seek app? If I remember right, someone in the original Seek Thermal thread mentioned that the app is packaged with the camera firmware and when the app is launched it will load the firmware onto the camera. If you haven't used the app with your original camera since you got it, it would be still running the oldest firmware, while the new cameras would be loaded with whatever was latest when they were produced. That would explain the difference in quality even if the manufacturing hasn't changed.

[IwuzBornanerd]

That is a dramatic improvement !

 I would have expected SEEK to make some announcement about releasing a new 'lower noise'' version of their microbolometer. I have not seen anything stating such.

Fraser

I too was surprised to not find anything about improvements to the "compact" on Seek's web site--not in any product description or any press release.  Maybe they don't want people to know about it lest it reduce sales of the "pro" version.

Have you connected your original 2015 Seek to a phone running the latest Seek app? If I remember right, someone in the original Seek Thermal thread mentioned that the app is packaged with the camera firmware and when the app is launched it will load the firmware onto the camera. If you haven't used the app with your original camera since you got it, it would be still running the oldest firmware, while the new cameras would be loaded with whatever was latest when they were produced. That would explain the difference in quality even if the manufacturing hasn't changed.

I don't have a phone to hook the thing to & I have not tried to find someone with a compatible phone who would be willing to register with Seek just to test my camera.  But if firmware alone would make the thing this much better wouldn't someone have already posted on here something like "Hey my camera suddenly makes much better images"?  I bought these 2 back in January so people have had plenty of time to notice.  I doubt firmware alone could make the whole difference as they would have to re-run the gain calibration for the camera.  They could read in the cal frames off the camera & scale them to the new 4 digit values but they would not be any more accurate than the old values so you'd see less improvement.

I have been thinking that people would not see the difference with the Seek app, or at least not the full potential improvement because Seek does not appear to have information stored on the camera to implement the bias correction--they would need another frame of data defining the pixel bias values used during the creation of the gain compensation factors.

Admittedly, though, the visible manufacturing changes are small.  The assembly number & revision level stamped on the circuit cards are identical in the old & new cameras, so I expect that all the parts are the same on both boards, unless they had a list of "selectable" part values.  This would not preclude any of the parts (such as the sensor) being a new revision, though.  The 2 notable visible changes are the thermal pads and a conductive rubber pin connecting the lid to the bottom of the case (see photos).

[Bud]

Maybe they just used a better lens ?

[IwuzBornanerd]

I used the lens (and lid) off of the old XR on the new non-XR to take this image of the back of the old XR board that I posted in the image gallery (in which you would see noise of 100mK), so I'd have to say the lens is not the difference https://www.eevblog.com/forum/thermal-imaging/thermal-imaging-gallery/?action=dlattach;attach=355569;image

Can someone explain, or point me to a web page/document that explains how a "better lens" could make such a profound difference in the amount of noise in the image?  Someone blamed the lens for @Bruno28's grainy images in the gallery also.  I can imagine that a wider aperture or less transmission loss would increase the signal-to-noise ratio but that would also change the calibration for temperature readings and probably also limit the sensing temperature range.  The lens aperture size is the same on the old & new XRs.  Not being educated on thermal lenses, I can pretty much only relate it to visible optics where the lens effects are mostly involved in image clarity & distortion, so I see the possible improvement in the thermal images as being in how crisp the image is rather than the noise level.

[Cat]

Very intresting discovery!

To test if the lower noise was from a different firmware or other hardware I want to show some results from my Seek Compact (non XR, non Pro) with different app-versions. The device was bought used and is already a few years old. Unfortunately the camera firmware did not change, probably because for my first test when I got it I already used the V2.1 app.

TL; DR: Noise performance was the same with both app versions.

Anyway here are the differences I noticed. Target was my (not so clean) desk and a baking oven.

Seek-app version 1.9.5.15:
Saves images with 832x624 px. ,the camera model in the EXIF is "PIR-206, 0.0.0.5"
The live image has "Seek thermal" displayed above the image.
Highest displayed temperature: +330°C
When very hot objects appear there seems to be an overflow and -40°C are displayed.

Seek-app version 2.1.1.3:
Saves images with 1280x960 px. , the camera model in the EXIF is "Compact"
Temperature indicators are smaller.
The live image has "Seek compact" as overlay in the image.
Highest displayed temperature: +324°C (maybe the heater turned off), same overflow error.
When very hot objects appear the automatic scale does not reset. So after viewing the oven the temperature scale was useless for room temperature (2.1.1_4.jpg)
Oh, and the registration can be skipped...

Because my Xperia Z3 has the micro-USB port in the left side and the Seek-app can not rotate the thermal image I used the app "Rotate Screen Orientation" from https://forum.xda-developers.com/showthread.php?t=2331514 to lock the screen orientation to landscape. Thus also the text in the images is 90° off...

Best regards,
Cat

[IwuzBornanerd]

Thank you @Cat for that report.  Your images look comparable to what I usually from a Seek, but I don't really know what I'm looking at since I don't know the full range of temperatures the palette is being applied to---as is often the case.  I have decided that the best way to compare cameras is to get an image with a full black-to-white gray scale palette spread over a narrow temperature range such as 3 degrees C--like @Bruno28 did with the grainy images he posted in the image gallery.  I say that because I know that my eyes can only clearly distinguish about 32 shades of gray, so if the image has the gray palette covering only 3 degrees C then I know I will be able to see a 0.1 degree C difference in temperature in the image.  It is harder to determine what I am seeing with a smooth color palette.  I have now coded that capability into my software so I will post some sample screenshots off of my PiZero.

The first image is from my new XR with my pixel averaging on, as in the images in my original post, with a 256 shade grayscale palette spread over a 3.2 degree C span.  The next image is the same but with the pixel averaging off.  This one should be good for an "apple to apple" comparison with @Bruno28's image of his speakers on page 4 of the image gallery (but I guess his is not all the way black to white)https://uploads.tapatalk-cdn.com/20170216/254e7ae5a01d631301fdc6190e36e352.jpg. The 3rd image is with the bias correction also off, so it is the noisiest one.  For comparison purposes, the last image is the OLD XR with pixel averaging & bias correction off, and the same 3.2 degree C temperature span.  I hope that using the gray scale palette this way can get all of us "on the same page", as the saying goes.

[IwuzBornanerd]

...
It is possible that SEEK and their microbolometer fabrication house have been making progress in taming the noise in their sensor. When first released, SEEK claimed that they would be developing and releasing a new microbolometer that did not need the Patent pixels to comply with FLIR patent claim. Maybe you are seeing a new microbolometer in your new camera ?
...
Fraser

It appears that the sensor is the same.  I began to think maybe this was the identification for only the logic chip the sensor is built on, but when I ran a search on EXC001 I got this: https://www.prnewswire.com/news-releases/seek-thermal-infrared-camera--raytheon-ir-microbolometer---reverse-costing-analysis-300037895.html and similar pages, as well as the github Seek hardware page (https://github.com/lod/seek-thermal-documentation/wiki/Seek-Hardware) identifying it as the Raytheon sensor in Seek's cameras.