For anyone who is impressed with the NETD figures provided by camera manufacturers, take them with a pinch of salt ! I personally could not believe a <20mK NETD from an uncooled camera in a real world scenario so decided to look into the T1020 specs. As has previously been discussed, where the NETD specification is concerned, all is not as it seems and the figures can be misleading. Manufacturers tend to create their NETD specification through careful laboratory testing setups and, in some cases, the use of specialist noise reduction algorithms that are not available to the end user ! See the linked paper for an interesting test of some high performance thermal cameras, including the FLIR cooled cameras and the T1020. In summary, the 'RAW' T1020 NETD turned out to be up around 110mK ! See page 74 of the test document for the plots in figure 4.11. How can FLIR claim a much lower NETD ? Well the FLIR Researcher 4 analysis software contains a noise reduction algorithm that can improve the apparent NETD through temporal frame averaging. Is this cheating ? Make up your own mind. The cooled cameras also exhibited higher NETD than claimed in the specifications but were far better than the Microbolometer camera figures. The cooled cameras did not have the option of applying the noise reduction algorithm however. In the testing the noise reduction was switched off as it could have caused the loss of wanted data.
FLIR are past masters of image improvement software in thermal cameras and it would appear that they are using their analysis software capabilities to improve the apparent NETD of some cameras but can you truly apply that NETD to the camera in its specifications without making it clear that it is a 'tweaked' figure ? Beware of the Marketing team 'tricks' ! It may, however, be fair to say that the T1020 has superior NETD to previous generations of camera, just not as amazing as claimed ! Note also the effects of integration time.
I reproduce the relevant text from the document below. The URL for the document is here:
https://core.ac.uk/download/pdf/250168454.pdfFraser
4.2 NETD values
In Figure 4.11, the histogram presents the distribution of the temporal standard
deviation values of the pixels calculated for every camera. An area of 100 × 100
pixels was selected for the analysis and thus the temporal NETD value is calculated
for 10 000 pixels over 32 sequential frames. In top of the figure one can see the
median of the standard deviations i.e. the temporal NETD of the camera.
Temporal NETD values of tested cameras.
The histograms present the distribution of standard deviations of each pixel in 100 × 100 area. The temporal NETD
of the camera is defined as the median value of all standard deviation values.
When the calculated temporal NETD values are compared to the NETD values provided by the manufacturers, one can notice that the calculated values based on the measurements are much higher. The noise reduction setting of the microbolometer
cameras was set off during the measurements. This is the feature of ResearchIR software that can be used with microbolometers. According to Danjoux (2019), the noise reduction feature performs some temporal frame averaging and thus smoothens
the output signal. According to discussion with Danjoux (2019), FLIR uses a special noise reduction filter for the specification measurements with microbolometers.
This filter is not identical with the feature of ResearchIR, but those are close to each other. Just for testing the noise reduction, one test was performed to A655sc camera where the feature was on. This caused the temporal NETD of A655sc to
drop to 28 mK. So this setting reduces the noise level significantly. Because the detected area was small and the heat effect was rapid, it was decided not to use the noise reduction feature in the camera tests to prevent reducing the desired signal as
well. For cooled photon detectors, there was not a noise reduction option, yet the calculated NETD values of the InSb cameras are higher than the values provided by the manufacturers. However, the NETD measurements of manufacturers are performed
in strictly controlled laboratory conditions with an object close to a perfect blackbody. It is understandable, that the values given by the manufacturers are smaller.
After all, one can reliably compare the NETD values of different cameras only if the test conditions and the signal processing algorithms are similar for every camera. When comparing the calculated NETD values of the cameras, one can notice that
the values are almost the same between microbolometers. Also the two InSb cameras manufactured by FLIR produce similar values. According to the calculated values, the InSb camera manufactured by Telops produces the best NETD value, 31.2 mK.
The standard deviation values of different pixels deviate more from the median value in microbolometers than in InSb cameras. Also in this respect, Telops is better than the other cameras.
The results of the spatial NETD calculations are also favourable to Telops as can be seen in Figure 4.12. Now the histogram presents the distribution of standard deviation values of 5 × 5 pixel areas that is calculated for every camera from the
averaged frame. The areas of the histograms for InSb X6900sc and MB T1020 are larger than others, as these cameras contain more pixels i.e. a higher resolution. Again, the standard deviation values between different 5 × 5 pixel areas vary more
with the microbolometers. The manufacturers do not provide spatial NETD value for the cameras.
The histograms present the distribution of standard deviations of 5 × 5 pixel areas. The spatial NETD of the camera is
defined as the median value of all standard deviation values. The total NETD values were also calculated for the cameras. The calculated temporal, spatial and total NETD values are presented in Table 4.2. It is important to notice, that the NETD is not an intrinsic feature of IR cameras. It will vary depending on the test conditions, the camera settings and the target temperature
as told in Subsection 2.2.6. Typically one desires to know what is the temperature difference that will create a signal equal to the noise level. After all, this reveals the magnitude of temperature differences that can be detected with the camera. Instead
of just checking the NETD value provided by the manufacturer, one should define the NETD value of the camera in real life operating conditions to acquire a realistic NETD value.
Table 4.2 The calculated temporal, spatial and total NETD values.
Cameras Temporal NETD (mK) / Spatial NETD (mK) / Total NETD (mK)
Telops Fast-IR 2K --- 31.2 / 6.1 / 31.8
FLIR SC7600 --- 44.6 / 8.8 / 45.5
FLIR X6900sc --- 48.4 / 10.9 / 49.6
FLIR A655sc --- 118.4 / 34.1 / 123.2
FLIR T1020 --- 110.3 / 31.3 / 114.6