If you’ve ever held a SWIR (Short-Wave Infrared) camera in our team’s hands—whether it’s for sorting recyclables, inspecting semiconductor wafers, or monitoring crop health—you know these devices aren’t your standard visible-light cameras. They work beyond what the human eye can see, capturing wavelengths between 1,000 and 2,500 nanometers, where materials like plastics, minerals, and biological tissues have unique spectral “fingerprints.” But for those fingerprints to be reliable, every SWIR imaging device from our facility goes through a calibration process that’s equal parts science, precision, and attention to small details. I’ve been leading calibration here for 12 years, and I still meet engineers who think “calibration” is just pushing a button. It’s not. It’s a ritual that ensures every pixel on your camera sees the world the way it’s designed to. Let’s break down how we do it, step by step, and why it matters for your work. SWIR Imaging
First, let’s get clear on what SWIR calibration actually is. Unlike visible-light cameras that mostly need white balance tweaks, a SWIR device has two core jobs to do well: capture consistent spectral data across its entire sensor, and map that data accurately to known material properties. If one pixel is slightly off, or the camera misreads the difference between a PET plastic bottle and a HDPE jug, your sorting line might contaminate a batch. If a semiconductor manufacturer uses an uncalibrated SWIR camera to check for micro-cracks in 300mm wafers, they might miss a flaw that costs thousands in rework. For us, calibration isn’t a post-factory step—it’s how we turn a good optical assembly into a tool you can trust.
We start calibration in a cleanroom, and it’s non-negotiable. Even a speck of dust the size of a grain of sand can block 1,550nm SWIR light (a common wavelength for industrial use) enough to throw off a pixel’s reading. The cleanroom is kept at a constant 22°C, with humidity held at 45% RH, because temperature changes expand optical components and alter their spectral response—something we learned the hard way when a heat spike during a 2019 calibration run made 17 cameras fail final testing. We also use nitrogen-purged workstations when handling unmounted sensors, to avoid water vapor in the air, which absorbs SWIR wavelengths and creates false signal noise. The first step here is always a visual and optical inspection: we check the lens for scratches or centering, confirm the sensor is seated evenly, and test the back-focus distance—critical because SWIR lenses work at different distances than visible-light lenses due to their longer wavelength.
Next, we do a dark current calibration. Dark current is the electrical signal generated by the sensor itself, even when no light is hitting it. For SWIR sensors (especially the InGaAs type we use in most of our industrial cameras), this signal is far higher than in visible sensors, and it increases with temperature—so ignoring it means your images will have a grainy, noisy background that hides real features. To calibrate dark current, we place the camera inside a light-tight enclosure that blocks all external light, then let it sit at the same temperature it will operate at during customer use (usually 20°C or 25°C, per their specs) for 30 minutes to stabilize. We then capture 100 successive frames, average them together, and subtract that average “dark frame” from every subsequent image the camera takes. For high-sensitivity models, we repeat this at three different operating temperatures (10°C, 25°C, and 40°C) to create a temperature-adjusted dark current lookup table. This is one of the steps that separates our calibration from a quick factory tweak—our tables ensure consistent performance whether a customer is operating in a cold warehouse or a hot outdoor field.
Once dark current is sorted, we move to spectral responsivity calibration, the heart of SWIR calibration. This is where we make sure every pixel responds the same way to light at every wavelength in the SWIR range (usually 900nm to 2,500nm for our cameras). A spectral reference source is key here—we don’t use cheap, broad-spectrum LEDs that vary in intensity; we use a NIST-traceable monochromator, which splits light from a calibrated tungsten-halogen lamp into discrete, precise wavelengths. The monochromator can produce light at every 10nm step between 900nm and 2,500nm, with an uncertainty of less than 0.5%—that’s the gold standard we rely on, because our customers need their measurements to be legally or operationally compliant.
To calibrate responsivity, we first capture a flat-field frame at each wavelength. A flat field is a uniform, diffused light source that illuminates every pixel on the sensor equally. For each wavelength, we take 50 frames (enough to reduce random noise), average them, then divide by the known intensity of the light coming from the monochromator. This gives us a responsivity value for every pixel at every wavelength. We then correct any pixels that are out of spec—old cameras might have a dead pixel here or there, so we create a bad-pixel map that replaces any faulty pixel with the average of its surrounding neighbors. For newer sensors, this step is less common, but we still do it, because even a 0.1% variance in a pixel’s response over time can add up to a big error over thousands of operating hours.
One thing we’ve learned over the years is that spectral responsivity isn’t enough on its own—we need to make sure the camera’s output is linear too. Linear response means that if we double the light intensity hitting the sensor, the camera’s signal doubles in return. If a camera has non-linear response, a customer might misclassify a material because the reflected light reading isn’t proportional to the material’s actual composition. To test linearity, we use the same monochromator, but instead of fixed intensity, we use neutral density filters that let us adjust the light intensity by 10%, 20%, up to 100% of our reference level. We capture frames at each intensity, then plot the signal vs. intensity for 10 random pixels across the sensor. If the R-squared value (a measure of linearity) is below 0.999, we adjust the camera’s gain settings during calibration. We’ve had customers come to us after buying a cheap SWIR camera that couldn’t tell the difference between clean and contaminated plastic—turns out the camera’s response was non-linear above 80% intensity, which is exactly what you get when sorting a fast-moving conveyor belt.
After spectral responsivity and linearity, we move to geometric calibration. SWIR cameras are often used for imaging large surfaces—from automotive parts to agricultural fields—so if straight lines in the real world look curved in the camera’s image, that’s a problem. For geometric calibration, we use a custom-made calibration target: a 1×1 meter aluminum sheet with 100 precision laser-etched dots, each spaced exactly 10cm apart, with a tolerance of 0.01mm. We mount this target at 5 different angles (0°, 15°, 30°, -15°, -30°) at a distance of 1 meter from the camera, then capture a frame at each angle. Using open-source and custom calibration software, we map the position of each etched dot in the camera’s image to its known real-world position, and create a distortion correction matrix. This matrix fixes both radial distortion (the curved edges you get from wide-angle lenses) and tangential distortion (the slight tilt of the lens that makes straight lines slant). For customers using our cameras for 3D imaging, we also add depth calibration here, using a secondary target with known depth markers to ensure accurate distance measurements.
But here’s the thing: calibration doesn’t stop when the camera leaves our facility. Over time, SWIR cameras get bumped, exposed to temperature changes, or have their lenses smudged, and their calibration drifts. That’s why we emphasize factory calibration as a baseline, and offer two types of on-site calibration for our customers: annual calibration (for industrial operations that run 24/7, like recycling sorting lines) and spot calibration (for customers who need a quick check after a lens change or sensor cleaning). For spot calibration, we’ve developed a self-contained calibration tool that’s the size of a shoebox—just plug it into your camera, run the software, and it adjusts the dark current and responsivity in 10 minutes, no external monochromator needed. Annual calibration is more in-depth, where we ship the camera back to our cleanroom and run through the full spectral, dark, and geometric checks, updating all lookup tables.
I get asked all the time: “Can’t I just calibrate the camera myself?” The short answer is, for most routine adjustments, maybe— but you’ll never get the same precision as a NIST-traceable calibration. We had a customer last year who tried to do his own spectral calibration for a semiconductor inspection camera, using a set of generic LEDs. He ended up with a 1.2% error in his micro-crack detection, which led to 5% of his wafers being misclassified as defective, costing him $20,000 in a single week. When he sent the camera back to us, we found his responsivity values were off by 8% at 1,310nm, the key wavelength for detecting wafer stress. It’s not that he didn’t know what he was doing—it’s that he didn’t have access to a calibrated light source with the precision needed for SWIR work.
Another common myth: “All SWIR cameras are calibrated the same way.” No way. A SWIR camera for fruit ripeness detection needs calibration at 900nm to 1,700nm, where sugars and starches reflect light differently. A SWIR camera for mineral sorting needs calibration at 1,500nm to 2,500nm, where silica, iron, and aluminum have distinct fingerprints. We tailor every calibration to the customer’s intended use, not just the camera’s specs. That’s why when we work with a new customer, we first ask for their application details, then adjust our calibration steps to match. For example, a customer sorting PET vs. PVC plastic doesn’t need calibration at 2,500nm, so we focus our responsivity checks on the 1,000nm to 1,700nm range, cutting calibration time by 20% and optimizing the camera for their specific needs.
Let’s wrap this up with a reminder of why this all matters. When you use a calibrated SWIR camera, you’re not just getting an image—you’re getting data you can trust. A recycler sorts more plastics correctly, a semiconductor plant produces fewer defective parts, a farmer identifies crop stress before it’s visible to the naked eye. That’s the work our team does, one calibration at a time. If you’re in the market for a SWIR imaging device, or you want to learn more about calibrating the one you already have, we’re here to help. We can walk you through the calibration process, tailor it to your application, and make sure your camera delivers the consistent, accurate data you need for your work.
Electro-optic POD References
- Dereniak, E. L., & Boreman, G. D. (2008). Infrared Detectors and Systems. John Wiley & Sons.
- InGaAs Photodetectors: Technology and Applications. (2021). Hamamatsu Photonics K.K. Technical Report.
- National Institute of Standards and Technology (NIST). (2020). Spectral Responsivity Calibration Procedures for Photodetectors. NIST Special Publication 250-86.
- Zhang, Z. (2000). A Flexible New Technique for Camera Calibration. IEEE Transactions on Pattern Analysis and Machine Intelligence, 22(11), 1330-1334.
Xi’an Zhongke Lead Ir-Tech Co., Ltd.
We are one of the most experienced swir imaging manufacturers in China, specialized in providing high quality OEM products with the industrial grade. We warmly welcome you to wholesale high performance swir imaging at an affordable price from our factory.
Address: Building 8,Hard Technology Enterprise Community No.3000,Biyuan 2nd Rd,High-Tech Zone Xi’an,Shaanxi,China
E-mail: sales@lead-ir.com
WebSite: https://www.leadinfrared.com/
