Machine Vision & Inspection

Machine Vision Lighting: Choosing LED Wavelength and Geometry for Automated Inspection

By Tech Led Updated Jul 23, 2026 9 min read

Machine vision lighting is the practice of selecting an LED light source (its wavelength and its geometry) to maximize the contrast a camera needs to detect a feature reliably. The wavelength is matched to the material and the defect: UV (365–405 nm) to excite fluorescence and reveal surface flaws, visible (450–660 nm) for color and general contrast, NIR (780–940 nm) to see through surface coatings and cut glare, and SWIR (1050–1650 nm) to read moisture, fill level, and material composition that visible light can't. The geometry (bright field, dark field, dome, backlight, or coaxial) controls how that light strikes the part. Getting both right is what separates a robust inspection from a flaky one.

Inspection goal Wavelength Geometry
Surface scratches, cracks, fluorescent markings UV 365 nm / 405 nm Dark field / low-angle
Color verification, print/label check Visible (450660 nm) Bright field / dome
Glare reduction, see-through coatings NIR (850940 nm) Diffuse / dome
Moisture, fill level, plastic ID, subsurface defects SWIR (1050–1650 nm) Backlight / diffuse
Edges, presence/absence, dimensional gauging Any (high contrast) Backlight (silhouette)

What machine vision lighting does

A machine vision system is only as good as the image its camera captures. Lighting determines whether a defect appears as a strong signal or disappears into noise. The objective is contrast: making the feature of interest look as different as possible from everything around it, consistently, regardless of ambient light or part-to-part variation.

Two levers create that contrast: the spectral lever (which wavelength interacts with the material in a useful way) and the geometric lever (the angle and diffusion of the light relative to the part and camera). LEDs are the dominant source for both because they offer selectable narrow-band wavelengths, instant strobing, stable output for repeatable measurements, long life, and compact form factors that fit ring lights, bars, domes, and backlights.

Choosing the wavelength

Different wavelengths reveal different things because materials absorb, reflect, and fluoresce differently across the spectrum.

  • UV (365–405 nm). Excites fluorescence in inks, adhesives, and biological residues, and rakes across a surface to expose scratches and texture. Used for verifying UV-cure adhesive coverage, detecting contamination, and surface-crack inspection. See the UV LED Guide and the fluorescence excitation guide.
  • Visible (450–660 nm). Blue, green, and red for color discrimination and general contrast. Monochromatic visible light is sharper than white for edge and feature detection; a wavelength is often chosen to contrast against a part's own color (e.g. red light darkens a green feature).
  • NIR (780–940 nm). Penetrates some surface coatings, reduces specular glare, and is invisible, so it doesn't interfere with operators or ambient color. Common for reading through tinted films, suppressing surface print, and high-contrast silhouette work.
  • SWIR (1050–1650 nm). Reads what silicon-based vision cannot: water content (1450 nm), hydrocarbons and plastics (1650 nm), fill level through opaque containers, and subsurface defects. Requires an InGaAs camera. See the SWIR LED Lighting Guide.

How wavelength selection creates defect contrast

Contrast is the gray-level difference between the defect and the background around it, normalized by that background. An inspection is only reliable when that gap stays wide across every part and every ambient condition. Wavelength is the first lever you reach for because a well-chosen band widens the gap before geometry, exposure, or algorithm ever touch the image. Four physical mechanisms do the work:

  • Differential reflectance and absorption. Materials reflect and absorb by wavelength. Pick the band where the defect and the good surface reflect most differently, and the defect separates from the background as raw signal rather than something the software has to rescue.
  • Color contrast (complementary wavelengths). A colored feature looks bright under light of its own color and dark under its complement. To make a colored defect stand out against a differently colored background, choose the wavelength that maximizes the reflectance difference between the two.
  • Fluorescence. UV and violet (365 to 405 nm) excite inks, adhesives, and biological residues that re-emit at longer visible wavelengths. Pair the excitation LED with a longpass filter on the camera so only the emission reaches the sensor.
  • Penetration depth. Short wavelengths scatter at the surface and expose texture and cracks; longer wavelengths penetrate and reveal subsurface voids or coatings. SWIR goes further and reads chemistry directly through molecular absorption bands (water near 1450 nm, hydrocarbons near 1650 nm).

The color-contrast rule is worth a table because it is the fastest win on any colored part. Same-color light brightens a feature; the complementary color darkens it.

Feature or defect color Light that brightens it (same color) Light that darkens it (complement)
Red Red (630–660 nm) Blue (450 nm) or green (520 nm)
Green Green (520 nm) Red (630 nm) or blue (450 nm)
Blue Blue (450 nm) Red (630 nm) or amber
Yellow Green (520 nm) or red (630 nm) Blue (450 nm)

A red defect on a green background is nearly invisible under white light because both look mid-gray to a monochrome sensor. Under blue light the green background stays bright and the red defect goes dark, and the contrast that was buried becomes obvious.

Matching wavelength to defect type

Working backward from the defect is usually faster than working forward from the catalog. The table below maps common inspection targets to the wavelength that separates them best and the geometry that delivers it.

Defect or feature Wavelength Contrast mechanism Typical geometry
Surface scratches, cracks, tool marks UV 365–405 nm or blue 450 nm short-wavelength scatter off surface texture dark field / low-angle
Fluorescent contamination (oils, adhesives, biological residue) UV 365 nm fluorescence emission flood + longpass filter
Color, print, and label defects visible monochromatic 450–660 nm reflectance and complementary contrast dome / bright field
Discoloration and stains on paper or film blue 450 nm or green 520 nm absorption differential dome
Glare on glossy or coated parts NIR 850–940 nm reduced specular reflection diffuse dome
Subsurface voids, delamination, adhesive gaps NIR 940 nm to SWIR 1200 nm penetration below the surface backlight / diffuse
Moisture, coating thickness, water ingress SWIR 1450 nm water absorption band backlight (InGaAs sensor)
Plastic type mix-ups, fill level in opaque containers SWIR 1650 nm hydrocarbon absorption backlight (InGaAs sensor)

Two practical rules keep the contrast you engineered. First, isolate the band: pair a narrow-band LED with a matched bandpass filter on the camera so ambient and mixed-color light are rejected and the measurement repeats part to part. Second, match the emitter count to the defect count. If one defect class dominates, a single wavelength is cheaper, cooler, and more robust; stations that must catch several unrelated defects often strobe two or three wavelengths in sequence and combine the frames.

Choosing the geometry

Geometry decides how the chosen wavelength interacts with the part's surface and reaches the camera.

Technique How it works Best for
Bright field Light reflects directly into the camera Flat, matte surfaces; general illumination
Dark field (low-angle) Light grazes the surface; only edges/texture scatter into the camera Scratches, engravings, embossing, edge defects
Dome (diffuse) Even, multi-angle light eliminates shadows and glare Curved, shiny, or specular parts
Backlight Light behind the part creates a silhouette Edges, presence/absence, dimensional gauging, holes
Coaxial (on-axis) Light injected along the camera axis via a beamsplitter Flat specular surfaces (wafers, glass, mirrors)

Strobing (pulsing the LED in sync with the camera) freezes motion on production lines and lets the LED run at higher peak intensity than continuous operation, improving signal without overheating.

Why LEDs for machine vision

  • Spectral precision. Narrow-band wavelengths selectable to the application, unlike broadband white sources.
  • Strobe capability. Microsecond pulsing synchronized to the camera freezes fast-moving parts.
  • Stability and lifetime. Constant output for repeatable measurements; 20,000–50,000 hour life cuts downtime versus fluorescent or halogen.
  • Form factor. Ring, bar, dome, backlight, and coaxial geometries in compact, integrable packages.

Tech-led supplies the LED emitters behind these systems across the full inspection spectrum, from UV through visible to IR/NIR and SWIR. For component selection, datasheets, and samples, contact Tech-led engineering.

Frequently asked questions

How do I choose the right wavelength for machine vision lighting?

Match the wavelength to the material and the feature. Use UV (365–405 nm) to excite fluorescence and reveal surface defects, visible light (450–660 nm) for color and general contrast, NIR (780–940 nm) to reduce glare and see through coatings, and SWIR (1050–1650 nm) to read moisture, fill level, or material composition. A useful rule: monochromatic light contrasting with the part's own color sharpens the feature.

How do I choose an LED for machine vision inspection?

Start from the defect, not the catalog. Identify the feature you must detect and the surface it sits on, then pick the wavelength that maximizes the reflectance difference between them: UV (365–405 nm) for fluorescence and surface cracks, monochromatic visible (450–660 nm) for color and print defects, NIR (850–940 nm) to cut glare and see through coatings, and SWIR (1450–1650 nm) for moisture and material composition. Then choose the geometry (dark field, dome, backlight, or coaxial) that delivers that wavelength to the part, and lock the result with a narrow-band LED plus a matched camera filter.

Which LED wavelength gives the best defect contrast?

There is no single best wavelength; the best one is the band where the defect and the background differ most in reflectance, absorption, or fluorescence. For a colored defect, the complementary color darkens it against a same-colored background. For surface texture, short wavelengths (UV and blue) scatter and expose it. For subsurface or moisture defects, NIR and SWIR penetrate or read absorption bands that visible light cannot.

Can a single LED wavelength handle all machine vision inspection tasks?

Rarely. One wavelength is ideal when a single defect class dominates, and it keeps the station cheaper and more repeatable. When a station must catch unrelated defects (a surface crack, a color error, and a moisture spot on the same part), integrators typically strobe two or three wavelengths in sequence and combine the frames rather than compromise on one broadband source.

What is the difference between bright field and dark field lighting?

Bright field directs light so it reflects straight into the camera, giving even illumination of flat matte surfaces. Dark field lights the part at a low grazing angle, so only edges, scratches, and texture scatter light back to the camera while flat areas stay dark, ideal for surface-defect and engraving inspection.

When should I use dome lighting?

Use a diffuse dome light for curved, shiny, or specular parts where direct lighting would create hotspots and shadows. The dome floods the part with even light from many angles, eliminating glare and giving uniform illumination for reliable inspection.

Why use infrared LEDs in machine vision?

NIR (780–940 nm) is invisible, so it doesn't interfere with operators or ambient color, it can penetrate some surface coatings and tinted films, and it suppresses specular glare. It's also useful for high-contrast silhouette and presence/absence checks.

What can SWIR lighting detect that visible light cannot?

SWIR (1050–1650 nm) reveals water content (around 1450 nm), plastics and hydrocarbons (around 1650 nm), fill level through opaque packaging, and subsurface defects, material properties that are invisible under visible or NIR light. SWIR imaging requires an InGaAs camera rather than a standard silicon sensor.

Why strobe machine vision lights?

Strobing pulses the LED in sync with the camera exposure to freeze motion on a moving line and to drive the LED at a higher peak intensity than continuous operation allows, increasing signal-to-noise without overheating the emitter.

Are LEDs better than fluorescent or halogen for machine vision?

Yes. LEDs offer selectable narrow-band wavelengths, microsecond strobing, stable repeatable output, 20,000–50,000 hour lifetimes, and compact geometries. Fluorescent and halogen sources are broadband, drift as they age, can't strobe cleanly, and have far shorter lives.

Specifying lighting for an automated inspection system? Contact Tech-led engineering for LED wavelength and component recommendations across UV, visible, NIR, and SWIR.

Share:
Tech Led

The Tech-LED team provides deep technical guidance on LEDs and optoelectronics, including design tips, industrial application notes, and updates on emerging wavelengths and packages.

Email