Noise Reduction on FLIR BOSON 640

[cnxunuo]

When I tried the built in NUC, it is so bad that each time the NUC is performed, a different line pattern showed up.
I decided to investigate this as it had never been this bad on any thermocamera i v ever seen.

50 frame is taken right after NUC is performed, with a piece of cardboard against the lens.
randomly distributed line-pattern looks horrible.


another 50 frames with a low contrast scene are taken sequentially after cardboard CAL
without any post processing, with built-in NUC,
image looks like this.


take a average of 50 frame, noise reduced as expected by statistics


subtracted my own NUC (50 frame of cardboard), noise reduced much drastically, but un-uniform line is still there/
this conclude the inadequate built-in NUC algorithm, and could potentially conclude the drifting gain of horizontal ROIC with a bit more detail.

[Hyper_Spectral]

Just out of curiosity, what temperature is the Boson at when you take these images? Have you compared say 10c vs 25c and 1-3 minutes after turn on vs 30 minutes after turn on?

Why does the image resolution appear so low on the final image of the 4 captures compared?

[bugi]

Why does the image resolution appear so low on the final image of the 4 captures compared?

The combined image seems to be just 2x zooms into the same spot, for easier comparison. To me the combined imaged doesn't look any more low resolution than what a zoom beyond 1:1 pixel ratio gives. At least downloading and comparing bigger image at 200% to combined image at 100% gives pretty much equal resolution and almost matching pixels (the small differences could be caused by JPG compression for each image etc.).

[cnxunuo]

There's never a noticable difference for any modern uncooled core running at 10 or 30degree c, only very antique uncooled FPA will produce a different in noise, but they usually have a TEC inside the package so in reality you can't notice them.

For compact core like this, unless been actively cooled which defeats it's small size, is impossible to run it at 30 degree c. When running at 10degree room temperature via USB, steady state chassis temperature usually rise above 55 degree.

[Fraser]

Worthy of note when discussing ambient/chassis temperature is what FLIR told me about the E4.

The E4 uses a non temperature stabilised microbolometer module. It does use a FFC flag to compensate for pixel drift however. When I asked FLIR about measurement accuracy as the cameras chassis rises above 30C, they were very honest. The camera is calibrated with a chassis temperature of 30C. As the chassis changes temperature, offset tables are applied to maintain measurement accuracy. These offset tables are generic and not specific to any one camera. The accuracy of the cameras measurements decreases as the chassis temperature gets further away from 30C. The Boson may well operate in the same way. This could be important if using a Boson to make measurements at different ambient temperatures.

Fraser

[cnxunuo]

From what I tried using an early E4 and what Mike's video suggested, the pcb mount sensor just compensates temp readout with a 'guessing” process, which saves them a tone of time during individual calibration, but spent the time only once during R&D, as their higher end units has a huge look-up table of target temp and chassis temp, may takes them hours on a huge rotary black body array and a thermal chamber to calibrate/generate.

just as a reference, its 17 indoor today, boson runs at 56 steady state with USB enabled.

[cnxunuo]

un fortunately in my case, brought from Flir China, they gave me a 9hz non-radiometric unit, with 2x the US price.
I will try using 16bit RAW to see if its actually compensated for chassis temp but just limited in software or most likely, 16bit vs target temp drift all over the place when chassis temp changes.

[Fraser]

Wow, that Boson is a hot little core  I was not expecting it to run that hot. Placing it inside a non ventilated casing could get interesting ! If you operate it in the tropics, at 30C ambient, I wonder how hot it would get then ? ~70C   

I see the spec sheet says the environmental operating temperature is -40C to +80C. If that is indeed ambient temperature, how hot must that little Boson get when in a +80C environment ?

Fraser

[cnxunuo]

A high temp “warning” dot will lite up above 105c
it's just an warning, not a overtemp shutdown,  i am curious about their design choices too....maybe it's just too small, needs a little conduction plate to somewhere else.

Their ASIC is so close to the sensor with a big board to board connector, I saw a constant drift to black (white hot) on upper right corner even after several hours of on-time, although it's not noticeable in normal contrast scene.

[cnxunuo]

Quick quantitative analysis:
turns out to be 87.98 raw count / degree K,
mine unit (16bit fake output, rescaled from 14bit) has a conversion fomula of
((16bit RAW)-20140/88) = Degree C.

This is done by pointing at black fabric tape covered hot water jug, 100, 55, and 16.8 degreeC, turns out to be very linear.

with this conversion rate, a randomly selected 12 pixel is graphed using matlab with Y axis in mK, and X axis in samples.

roughly 100mk Peak-peak, 40mk RMS value, which falls into the <40mk datasheet value.

[Hyper_Spectral]

A high temp “warning” dot will lite up above 105c
it's just an warning, not a overtemp shutdown,  i am curious about their design choices too....maybe it's just too small, needs a little conduction plate to somewhere else.

Their ASIC is so close to the sensor with a big board to board connector, I saw a constant drift to black (white hot) on upper right corner even after several hours of on-time, although it's not noticeable in normal contrast scene.

Hmm. I feel like the Boson was designed specifically for low SWAP performance and as such they expected the core to be in a highly ventilated area like a gimbal hanging off a drone. Those temps are way higher than I expected.

The tau 2 core I'm familiar with, and the WIRIS cores, generally needed time to warm up in <50f temps for best performance. I figured the boson would be similar

Nonetheless, the noise from that boson is absurd. I expected more from FLIR