NIR Imaging for Machine Vision: 850nm vs 940nm, Bandpass Filters, and IR-Corrected Lenses

850nm vs 940nm: how to choose the right NIR wavelength

850nm and 940nm are the two standard near-infrared wavelengths for machine vision illumination. The choice affects three things at once: how much of the emitted light the sensor actually converts to signal, whether the illuminator is visible to a human observer, and which bandpass filter the system needs downstream. Pick deliberately based on the deployment constraints, not by defaulting to whichever wavelength a vendor happens to stock.

850nm: higher sensor QE with a faint visible glow

Silicon CMOS sensors have higher quantum efficiency at 850nm than at 940nm. For the same illuminator drive power, an 850nm source typically delivers more usable signal at the sensor, which supports shorter exposure times or lower illumination power for a given signal level. The tradeoff is that 850nm LEDs emit a faint visible red glow. Human sensitivity in the deep red falls off steeply but does not fully vanish until roughly 750nm, and the short-wavelength tail of an 850nm LED's broad emission puts a small fraction of its output near that edge. At high drive currents, this glow is detectable in dark environments. Where appearance standards or nearby human observers require no visible LED indication, 850nm may not be acceptable without reducing drive current.

940nm: no visible glow but roughly half the silicon QE

940nm illumination sits far enough outside the eye's visual response that it produces no perceptible glow under normal operating conditions. The cost is sensor response: silicon quantum efficiency at 940nm is commonly on the order of half or less of its value at 850nm, and the exact ratio varies by sensor design. That gap means a 940nm system typically needs more illumination power, a longer exposure time, or both to reach the same signal-to-noise ratio an 850nm system achieves at the same range. High-throughput lines where exposure time is a binding constraint should verify this margin against the actual sensor QE curve before committing to 940nm.

Atmospheric and outdoor considerations

Outdoor NIR systems face an additional variable: ambient sunlight contains strong NIR content that competes with active illumination. The solar spectrum is not flat across the NIR band, though. Atmospheric water vapor absorbs solar radiation in a band around roughly 930nm-970nm, which measurably suppresses ground-level solar irradiance at 940nm relative to 850nm. This is a large part of why 940nm is a common choice for outdoor active-illumination systems such as driver-monitoring cameras, face recognition, and eye-tracking: the ambient solar background a 940nm sensor competes against is genuinely lower, partially offsetting the sensor QE penalty covered above. 850nm sits outside that absorption band and sees the full, largely unattenuated solar NIR floor. A narrow-passband bandpass filter matched to the illumination wavelength rejects most of the ambient contribution at either wavelength, but at long range or in bright sun the residual ambient NIR floor can still reduce contrast, and 940nm carries a smaller residual floor than 850nm precisely because of the atmospheric water-vapor absorption. The mitigation for what remains is a well-matched narrow filter and, where needed, synchronized strobing of the illuminator with a short exposure window rather than a wavelength change.

Range also depends on the same throughput gap that drives the 850nm-versus-940nm exposure margin. Because 850nm delivers more effective photons to the sensor per watt of illuminator power, 850nm systems typically reach longer detection ranges than 940nm systems at the same illuminator size and power budget. A 940nm unit can be built to reach comparable range, but typically only by adding LEDs or raising drive current, which means a larger, hotter, higher-current illuminator. When comparing illuminator datasheets, check LED count and drive current rather than trusting a stated range figure at face value.

Filter architecture follows directly from the wavelength decision: a system built around 850nm pairs with a CBP850 single-bandpass filter or a CDB850 dual-bandpass filter for passive day/night operation, while a 940nm system uses the CBP940 or CDB941 equivalents (see the table above for links). These are not interchangeable: a 940nm bandpass filter blocks 850nm illumination and will produce a dark image if paired incorrectly. See bandpass filter machine vision for how center wavelength and passband width factor into that selection.

Why NIR reveals contrast that visible light misses

NIR imaging is useful because reflectance, transmission, and absorption vary with wavelength, and many materials look different in the NIR band than under white light. That wavelength-dependent difference is the entire basis for NIR machine vision. It is not a general-purpose upgrade to image quality.

Inks and printed graphics

Carbon-based inks common in printing absorb visible light and appear dark, and that absorption carries through into the NIR band. Carbon black is a broadband absorber, so carbon-ink printing stays dark and keeps its contrast against the substrate under 850nm or 940nm illumination. Dye-based inks behave differently: many dyes that look opaque black to the eye are largely transparent in NIR, so a barcode or graphic printed in dye-based black ink can effectively disappear in a 940nm image, revealing substrate texture beneath it. Some specialty inks that look identical to the eye produce measurable NIR contrast differences, which supports authentication and sorting applications visible imaging cannot.

This same property runs in the opposite direction for OCR and print-verification systems. If the goal is reading or verifying the printed characters themselves, and those characters are printed in dye-based ink, NIR illumination can wash out the very feature the system needs to detect. Carbon-based ink does not have this problem; it stays dark and legible under NIR. The fix is not to avoid NIR outright but to verify ink composition on the actual production substrate before committing to a wavelength, since dye-based and carbon-based inks do not behave identically in the NIR band.

Surface glare and specular reflections

Specular reflections from glossy or metallic surfaces saturate visible pixels when ambient lighting is not tightly controlled. Narrow-band NIR illumination paired with a matching bandpass filter rejects the broadband ambient visible glare, so the sensor mainly captures the NIR contribution scattered from the surface. This does not remove specular reflection from the NIR source itself, but it removes the visible-spectrum glare that dominates in typical factory lighting. See stray light in machine vision for a broader treatment of glare control.

This is a different mechanism than reducing glare with polarizing filters or diffuse lighting geometry, which work in the visible band by managing the angle or polarization state of the reflected light rather than its wavelength. NIR bandpass rejection and polarization control are not mutually exclusive; some inspection stations combine an NIR bandpass filter with a polarizer on the illuminator to address both broadband ambient glare and angle-dependent specular hotspots in a single setup.

Coatings, films, and biological materials

Thin films and surface coatings that look uniform under white light can show different NIR reflectance depending on composition or thickness. Biological and food-grade materials often show NIR contrast tied to water content or organic compound concentration, which is why NIR shows up in fill-level inspection, food sorting, and pharmaceutical packaging verification. A liquid level inside an opaque plastic bottle, for example, can sometimes be distinguished from the empty headspace above it in NIR even though both look identical from outside under visible light, because water absorbs more strongly in parts of the NIR band than the surrounding plastic or air.

What NIR does not do

NIR imaging is not universally better than visible imaging. Tasks needing color discrimination or legibility of visible ink are usually better served by visible light. NIR does not penetrate most opaque materials the way X-ray does, and useful penetration depth in paper, plastic, or film varies by material and must be validated on real samples. A materials datasheet showing an NIR contrast difference does not guarantee the same result on the actual parts or packaging in production.

Filters, lenses, and switching architectures for NIR systems

Beyond sensor and illumination choice, an NIR system requires three hardware decisions: filter type, lens IR correction, and whether the system needs to switch between visible and NIR modes at all.

Filter selection

The IR-cut filter is the most commonly misunderstood element. It blocks NIR and must be out of the optical path for NIR imaging to function at all, but removing it is necessary, not sufficient. Without a bandpass filter, the sensor also sees ambient visible light, which degrades contrast and signal-to-noise ratio.

A single bandpass filter at 850nm or 940nm passes only the illumination band and blocks visible light, which is the standard configuration for pure NIR inspection. A dual-bandpass filter passes a visible window and a defined NIR window simultaneously in one fixed element, used when the system must capture color and NIR in the same frame on an RGBIR sensor; it cannot fully reject either band and does not separate them in time.

An electronic switcher, such as the CLA216-ICR-850BP shown below, carries an IR-cut filter and an 850nm bandpass filter and moves one into the optical path on command, giving true day/night separation with full rejection in each mode. See bandpass filter machine vision for CWL and FWHM selection detail.

When lens IR correction matters

If a system operates only in NIR, focus the lens at the NIR wavelength and IR correction is unnecessary. If the system must produce sharp images at both visible and NIR wavelengths from the same mechanical focus position (day/night cameras, RGBIR sensors, or dual-mode inspection), an IR-corrected lens is required. A standard lens has chromatic focus shift between visible and NIR because of dispersion in the glass; removing the IR-cut filter does not close that gap, since the shift originates in the lens optical design, not the filter. An IR-corrected lens brings the visible and NIR focal planes close enough together to satisfy the system's resolution requirement across both.

Sensor selection for NIR systems

Not every sensor is a good NIR candidate. Many consumer-oriented and some machine vision sensors have an NIR-blocking layer built into the pixel stack itself, independent of any external IR-cut filter, which caps usable sensitivity at both 850nm and 940nm regardless of what filters the system uses. Before committing to a wavelength, pull the sensor's spectral quantum efficiency curve and read the actual value at 850nm and at 940nm rather than assuming a generic silicon response. See image sensor selection for machine vision for how NIR response factors into the broader sensor decision alongside resolution, pixel size, and shutter type.

RGBIR sensors are a special case: they carry a per-pixel filter pattern that lets some pixels respond primarily to NIR while others carry RGB color information, enabling simultaneous color and NIR capture from a single sensor without a mechanical switcher. These sensors depend on a dual-bandpass filter in front of the lens to define the NIR passband cleanly; without it, NIR leakage into the RGB channels degrades color accuracy in daylight.

Chief ray angle and filter placement

Interference bandpass filters shift their center wavelength toward shorter wavelengths as the angle of incidence increases. In compact M12 lens assemblies where the filter sits close to the sensor, chief ray angles at the image corners can reach 15 degrees to 20 degrees. A filter specified for 850nm at normal incidence will show a modest blue-shift in effective center wavelength at the corners. For most machine vision bandpass filters this shift is small relative to the passband width, but it is worth checking against the filter's angle-of-incidence spec if the system uses an unusually narrow passband or a wide-angle lens. See lens chief ray angle and mismatch for the underlying geometry.

NIR system validation checklist

Run this checklist at the real working distance, aperture, and exposure conditions of the deployment. Bench tests at arbitrary conditions routinely fail to predict production performance in NIR systems, because ambient NIR content, filter angle-of-incidence shift, and sensor NIR response all vary between a lab bench and a production line in ways that a quick prototype test will not surface.

• Confirm the IR-cut filter is absent. Remove it physically or verify the camera module was built without one.
• Verify the illumination wavelength. Measure with a spectrometer if the emitter's peak wavelength is not specified on the datasheet.
• Match the bandpass filter to the illumination source. A 20nm mismatch can reduce throughput by 50% or more on a narrow-passband interference filter.
• Capture NIR on and off at the target scene. The on/off contrast ratio on the target feature confirms whether active illumination dominates ambient NIR content.
• Measure signal-to-noise ratio on the real target feature at the operating exposure time and confirm it exceeds the detection algorithm's minimum.
• Check focus stability across wavelengths. If the system also images in visible light, set focus under visible illumination, switch to NIR without adjusting focus, and compare sharpness at the image center and corners on a resolution chart.
• Test with representative target materials, not only calibration targets. NIR reflectance differences from datasheets may not match actual parts or packaging.
• Check behavior at temperature extremes. LED peak wavelength and lens focus can shift with temperature and degrade NIR performance at hot or cold conditions.
• Check illuminator uniformity across the full field of view, not just at the center of the image. NIR illuminators typically fall off toward the edge of their beam angle, and a lens with a wider field of view than the illuminator's effective beam angle will show a dark corner or vignette that has nothing to do with the lens itself. Confirm the illuminator's beam angle equals or exceeds the lens field of view at the intended working distance, and measure signal-to-noise ratio at the corners of the frame, not only at the center, since corner performance is usually the first thing to fail as range increases.
• Re-run validation after any bill-of-materials change. Swapping an LED supplier, a filter vendor, or a sensor revision can shift peak wavelength, quantum efficiency, or filter center wavelength enough to invalidate a prior validation pass, even when every part number on the drawing looks unchanged.
• Document the validated configuration. Record the illumination wavelength, bandpass filter part number, lens model, working distance, aperture, and exposure time. Any change requires re-validation.