As an Amazon Associate, Akermin earns from qualifying purchases. Product links on this page are affiliate links.
NETD stands for noise equivalent temperature difference: the smallest temperature difference a thermal camera can distinguish from its own electronic noise, quoted in millikelvin. A camera specified at 40 mK can separate a surface four hundredths of a degree warmer than its surroundings from random fluctuation, and one at 100 mK cannot. Resolution decides how finely a feature is drawn and how small an object can fill enough pixels to be judged, so for insulation, damp and draught work, where the subject is large and the differences are faint, sensitivity is the figure that decides whether the problem appears on screen at all.
NETD in one sentence, then in detail
Every detector produces a fluctuating output even when pointed at a perfectly uniform surface. NETD is the scene temperature difference that produces a signal equal to the size of that fluctuation. It is a noise floor expressed in temperature units, which is why it is quoted in millikelvin: 40 mK is 0.04 degrees Celsius, and 1,000 mK is one full degree.
Three details stop the number being misused. First, lower is better, and the scale is not linear in usefulness: moving from 100 mK to 50 mK transforms what you can see on a wall, while moving from 40 mK to 35 mK rarely changes a decision. Second, the figure is only meaningful with its conditions, since it is measured at a stated scene temperature, aperture and frame rate; a slower lens or a colder scene worsens it. Third, it is not accuracy. A camera can resolve a four hundredths of a degree difference between two patches while being two full degrees wrong about what either patch actually is.
Resolution tells you where, sensitivity tells you whether
Resolution is spatial sampling. Divide the field of view by the pixel count across it and you get the angle each pixel subtends, and from that the area each pixel covers at a working distance. With a 25 degree horizontal field of view spread across 256 pixels, each pixel covers about 0.1 degrees, which is roughly 1.7 milliradians, which at 3 metres is a patch about 5 mm across. Reliable temperature measurement needs the target to fill several pixels, not one, so that camera needs an object of roughly 15 mm at 3 metres before the reading is trustworthy.
Put two jobs against that arithmetic and the priority inverts between them. A damp patch is 300 mm across and perhaps 0.3 degrees cooler than the wall around it, so it fills sixty pixels across and is trivially resolved, while the temperature difference sits only a few times above a 100 mK noise floor. That job is sensitivity limited. A loose terminal in a consumer unit is 5 mm across and 30 degrees hotter than its neighbours, so it is a single pixel at 3 metres and enormously above any noise floor. That job is resolution limited, and the answer is to get closer or use a camera with a longer lens or more pixels. Our pages on thermal camera resolution and comparing thermal camera resolution work through the pixel geometry in more depth.
Reading mK figures across four listings
Sensitivity is published more often than it used to be, and it appears under several names: thermal sensitivity, NETD, or a phrase about detecting differences as small as a given temperature. All three mean the same quantity. Here is what four current listings state.
| Camera | Stated sensitivity | Native detector | Output resolution | Stated accuracy |
|---|---|---|---|---|
| DCTOPT handheld | Below 50 mK | Not stated in the listing bullets | 240 by 240 super resolution in the bullets, 240 by 180 in the title | 2 degrees Celsius or 2 percent |
| Thermal Master P3 phone attachment | 0.035 degrees Celsius, equal to 35 mK | 256 by 192 | 512 by 384 upscaled | Plus or minus 2 degrees Celsius |
| REVASRI phone attachment | 40 mK | Not stated in the listing bullets | 320 by 240 super resolution | Plus or minus 2 degrees Celsius or 2 percent |
| Thermal Master Thor002 handheld | 40 mK | 256 by 192 | 512 by 384 upscaled | Plus or minus 3.6 degrees Fahrenheit |
Two things stand out. The published sensitivity figures are clustered tightly, from 35 mK to below 50 mK, so sensitivity is no longer the thing that separates these units. And the resolution column is a mess of different quantities presented as one. The DCTOPT thermal camera is the clearest example: its bullets state a 2.8 inch display with 240 by 240 super resolution technology and a thermal sensitivity below 50 mK, while its own title states 320 by 240 for the screen and 240 by 180 for super IR resolution. Those cannot all describe the same thing, and the native detector format is not stated in the listing bullets at all. The REVASRI imager is in the same position: it states 320 by 240 super resolution with a 25 Hz refresh and 40 mK sensitivity, and leaves the underlying detector format unstated.
Super resolution is not sensitivity
Two of these listings are refreshingly explicit about what their big pixel number is. The Thermal Master P3 states that its processing upscales a native 256 by 192 thermal resolution to 512 by 384, and the Thermal Master Thor002 states the same path from 256 by 192 to 512 by 384. That is a real detector of 49,152 elements presented as an image of 196,608 pixels. Three pixels in four are computed from their neighbours.
Interpolation has genuine uses. It smooths the blocky edges that make a low resolution thermal image hard to interpret, and it makes on screen cursors and measurement boxes easier to place. What it cannot do is add thermal information, because the detector never sampled it. A difference that fell below the noise floor is absent from the frames the upscaler receives, so it stays absent afterwards. This is why NETD and native format are the two figures to compare, and the upscaled number is a presentation choice.
A third resolution lurks in the same listings and belongs to neither category. The Thor002 states a 3.5 inch 640 by 480 IPS display alongside a 2 megapixel visible light camera. Panel pixels and visible camera pixels have nothing to do with thermal detail. When you see four resolution figures on one page, sort them into detector, upscaled output, display panel and visible camera before comparing anything. Our list of thermal cameras with low NETD records which listings publish each of them.
Why accuracy, sensitivity and resolution are three separate numbers
These three get conflated constantly, and the consequences show up in reports. Sensitivity is about differences within one image. Resolution is about how small a thing can be. Accuracy is about the absolute scale, and it is the weakest of the three on every handheld camera, typically plus or minus 2 degrees Celsius, because converting radiated energy into a temperature requires knowing the surface emissivity, the reflected background, the distance and the atmospheric path.
- A camera at 35 mK with plus or minus 2 degrees of accuracy can tell you one patch is 0.1 degrees cooler than another with confidence, and cannot tell you either absolute value to better than 2 degrees.
- That is why thermography is comparative. You work in temperature differences against a reference area of the same material, and the accuracy term largely cancels.
- Emissivity adjustment matters more than either headline figure for absolute work. Shiny metal radiates poorly and reflects its surroundings, so an uncorrected reading on a bus bar can be out by tens of degrees, which no sensitivity figure rescues.
- Minimum focus distance is the forgotten fourth number for close work. The REVASRI 320×240 states a manual focus lens with a minimum focusing distance of 8 mm, which is what makes circuit board inspection possible at all.
If you only ever need to know which of two things is warmer, sensitivity and resolution are your figures. If a number is going into a report that someone else will act on, accuracy, emissivity handling and a documented reference method matter more than either. Our comparison of a thermal camera against an infrared thermometer covers the same split for spot measurement, where a single averaged circle replaces the image.
Choosing the figure that matches your inspection
Decide which limit binds your work before comparing any two cameras. Large subject, small difference means sensitivity first: insulation gaps, cold bridging, draughts, damp, underfloor heating runs, radiant floors. Small subject, large difference means resolution, range and focus first: terminals, breakers, bearings, motors, exhaust components, circuit boards.
Then check the figure is stated rather than implied. A sensitivity figure with no conditions is weaker than one quoted at a stated scene temperature, a native detector format is worth more than an upscaled one, and an emissivity setting is worth more than either if absolute numbers are the point. For the definition on its own and the way different manufacturers phrase it, our page on NETD on a thermal camera keeps the reference version, and the figures in the table above show how rarely the complete set appears in one place.
Frequently Asked Questions
What does NETD stand for?
Noise equivalent temperature difference. It is the temperature difference in the scene that produces a signal equal to the noise of the camera itself, so it marks the floor below which a real difference is indistinguishable from random fluctuation. Lower is better, and it is quoted in millikelvin, where 1,000 millikelvin is one degree.
Is 40 mK a good figure?
For handheld work it is a strong figure, and it is what several current pocket and phone attachment cameras publish. Below about 50 millikelvin you can see the faint patterns that matter in insulation and moisture work. Around 100 millikelvin those patterns start disappearing into noise, which is why older or cheaper sensors struggle with cold bridging.
Does NETD change with the lens or the scene?
Yes, which is why a bare figure is only half a specification. NETD is quoted at a stated scene temperature, usually around 30 degrees Celsius, with a stated aperture and frame rate. A slower lens passes less energy and worsens the effective figure, and averaging several frames improves it at the cost of lag on a moving target.
Does super resolution improve NETD?
No. Upscaling works on the frames the detector has already produced, so it cannot recover a difference the detector never resolved from its noise. Sensitivity is a property of the detector and its electronics, and the only thing interpolation changes is how smooth the picture looks.
Why can a camera see a 0.04 degree difference but only measure to plus or minus 2 degrees?
Because sensitivity and accuracy are different quantities. Sensitivity is about distinguishing differences within one image; accuracy is about anchoring those differences to an absolute scale, which depends on emissivity, reflected background, distance and calibration. This is why thermography is done with difference measurements and reference areas rather than absolute readings.
Which figure should decide my purchase?
Whichever matches the dominant limit in your work. Insulation, draughts, damp and underfloor heating involve large areas with small temperature differences, so sensitivity leads. Electrical and mechanical faults involve small objects with large differences, so resolution, minimum focus distance and the top of the measurement range lead.