SWIR camera applications: pick the band by what the material absorbs

A short-wave infrared camera does not buy resolution or speed. It buys contrast at wavelengths where materials behave differently than they do under visible light, and that single idea decides every SWIR project. Choose the wavelength first, from what the material absorbs, transmits or scatters, then size the camera, lens and illumination around it. This guide maps the common industrial cases to the band each one needs: moisture in food, paper and coatings at 1450 nm, polymer sorting across the long half of the band, subsurface wafer checks through silicon above 1100 nm. It works the optics arithmetic for a real conveyor line, compares the lenses and lights that transmit past 900 nm, and marks the jobs where a standard CMOS camera or a plain 850 nm NIR setup still wins.

The decision rule: name the dark and the bright

SWIR earns its cost in exactly one situation: you can complete the sentence 'X must read dark while Y reads bright at λ nm', and the difference comes from absorption rather than geometry. Water against dry dough at 1450 nm. One polymer against another at its overtone band. A metal bond against bare silicon above 1100 nm. If you cannot fill in X, Y and the wavelength, you do not yet have a camera decision — you have a filter and lighting test to run first.

Contrast that lives in shape, height, colour or printed text stays on standard CMOS hardware with conventional lighting geometry. SWIR vs NIR vs visible imaging: which band to inspect with walks the full band comparison; this guide takes the cases that land beyond 900 nm.

Three spectral facts that drive the band choice

Water absorbs in discrete peaks. Liquid water has absorption bands near 970, 1190, 1450 and 1940 nm. An InGaAs camera covers 900-1700 nm, so three of the four sit within reach and the strongest, 1940 nm, does not. That is why 1450 nm became the working standard for moisture imaging: illuminate around that line, and wet areas read dark while dry surroundings stay bright, with contrast deepening as moisture rises. It is also why the boundary between NIR and SWIR is commonly drawn here.

Silicon turns transparent above about 1100 nm. A wafer is opaque in visible light and grows increasingly clear with wavelength, so a SWIR camera images structures, bond interfaces and defects beneath the surface. Transmission improves with wavelength and with thinner material, so usable depth depends on wafer thickness and doping. Design a through-wafer illumination path and prove the contrast on your own wafer specification.

Plastics carry distinct overtone bands. Across the long-wave half of the band, each common polymer keeps its own absorption signature, so PET, PE, PP, PVC and PS — identical to the eye — separate into distinguishable grey levels with two or three bandpass images. The same logic pulls organic contaminants out of a food stream against a good-product background.

A fourth effect helps on dirty lines. Elastic scattering in haze, smoke and dust falls roughly with the fourth power of wavelength, so moving illumination from 775 nm to 1550 nm cuts scattering about sixteenfold. Near-infrared wavelength selection for water and silicon inspection works through these two material systems band by band.

The application classes and the band each one needs

Read the table from the middle column first: the band is the purchase. Camera and light become straightforward once the wavelength is fixed.

The last row is there on purpose. If the variable you want is temperature, an infrared camera in the 8-14 μm LWIR band is the correct instrument, and no SWIR setup substitutes for it.

Worked example: moisture streaks on a bakery belt

The task: wet streaks at least 0.5 mm wide on a belt moving 120 mm/s, imaged so that wet reads dark against dry dough. This is the classic 1450 nm job, and the arithmetic below generalises to any absorption-contrast station.

Start from the camera. The LS-IRC-0029 is a 0.3 MP SWIR area scan camera with 15 μm × 15 μm pixels, a global shutter, a C-mount and a GigE interface on a 1/1.2-inch format. For the arithmetic, take the common 640 × 512 array of this resolution class: 640 × 15 μm gives a 9.6 mm sensor width, 512 × 15 μm a 7.68 mm height, and the diagonal works out to 12.3 mm.

Choose the lens. The LS-SWL-0229 is a 35 mm, F1.4 C-mount lens corrected for 900-1700 nm. Its 11-16 mm image circle clears the 12.3 mm diagonal of the example array.

At a 600 mm working distance:

• m = 35 / (600 - 35) = 35 / 565 = 0.062
• pixel on the belt = 15 μm / 0.062 = 242 μm ≈ 0.24 mm
• field of view = 9.6 / 0.062 = 155 mm wide × 7.68 / 0.062 = 124 mm high
• streak resolution = 0.5 mm / 0.242 mm = 2.1 pixels
• exposure cap = 0.242 mm / 120 mm/s = 0.002 s ≈ 2 ms for one-pixel blur

A tighter cell changes the lens, not the method: a 25 mm SWIR lens reaches the same 155 mm field at 428 mm working distance (m = 25 / 403 = 0.062).

The 1450 nm filter sets the light budget. Fit a narrow 1450 nm bandpass in front of the lens, illuminate broadly across the band, and wet areas fall to near-black on a bright dry background. Two milliseconds is a long exposure by machine-vision standards, which is why uncooled InGaAs runs this class of line with margin to spare. The order for the station: a 0.3 MP InGaAs camera, a 35 mm SWIR lens at about 600 mm working distance, a 1450 nm bandpass, broadband SWIR illumination, and a 2 ms exposure cap.

Picking the camera: resolution, cooling, area or line

InGaAs is the sensor technology behind industrial SWIR, because silicon CMOS response fades out near 1000-1100 nm, right where the useful bands begin. Three selection calls matter more than the rest.

Resolution. The line-side classes are 0.3 and 1.3 MP. The Hikrobot MV-CI013-GS-NN is a fanless 1.3 MP SWIR camera with an IMX990 sensor, a 1/2-inch format, a C-mount and a GigE interface — the class to choose when a wet streak or a wafer feature must be measured, not merely detected. The 640-column class detects fine, as the worked example shows, and reads out faster over the same GigE link.

Cooling. Thermoelectric cooling suppresses dark signal for long exposures and passive scenes. With active illumination and millisecond exposures, an uncooled, fanless camera has ample margin and copes better with cabinet heat. Uncooled SWIR camera stability for inspection systems shows what drift to expect and how to qualify a camera for it.

Area or line. Area scan suits cells, wafers and bins. Line scan suits webs and free-fall sorting, where the belt supplies the second dimension and uniform illumination along the scan line matters more than raw sensitivity. Sorters commonly strobe the light in step with the camera to hold exposure short.

Both cameras speak GigE, so either drops into a standard GigE Vision acquisition chain without frame-grabber hardware. The cameras topic page collects the rest of the camera-selection guides on this site.

Lenses and lights that transmit past 900 nm

Ordinary visible-light lenses quietly fail here: glasses and coatings optimised for 400-700 nm lose transmission across the SWIR band and shift focus, so a lens focused under white light is out of focus under SWIR. SWIR lens and UV lens basics: what changes outside visible light explains the optical differences. Select from the thermal, SWIR and UV lenses category by wavelength range and image circle, matching the circle to the sensor diagonal: 8 mm for 1/2-inch, about 11 mm for 2/3-inch, 16 mm for 1-inch.

If the same camera must also focus under visible light during set-up, the LS-SWL-0246 covers 400-1700 nm at F1.8 in a 25 mm focal length for a 1/2-inch sensor. For larger formats, the LS-SWL-0207 is a 25 mm, 5 MP SWIR lens for 2/3-inch sensors, and the 35 mm lens above covers 1-inch formats.

All 3 share: Mount C-mount. Values as printed in each product's data sheet; open a part for its drawing and full specifications.

On hot lines, focus drifts as the lens temperature moves; infrared lens athermalization: keeping focus as temperature changes quantifies the drift and the standard fixes.

Illumination splits into two jobs. General NIR contrast — fill level through dark plastic, print on brown packaging — runs on the 850-856 nm class of infrared LED units listed under special lights; NIR lighting for fill level inspection through packaging shows that set-up end to end.

Moisture work needs the 1450 nm line: a narrow bandpass on the camera and a strong broadband or 1450 nm-class source doing the work. Tighter filters raise contrast and discard signal, so budget illumination power as the filter narrows. The light step of the configurator narrows the catalogue by wavelength, and if contrast still disappoints after filtering, machine vision stray light: diagnose low image contrast gives the diagnostic order.

When SWIR is the wrong tool

Three everyday cases stay outside SWIR. Colour and print defects: a standard CMOS camera with the right visible-lighting geometry sees them better than any infrared band will. Heights, warpage and dents: those belong to dark-field lighting or 3D profilometry, where the contrast is geometric. Temperature: when the number you want is degrees, use a thermal camera — thermal camera NETD and lens choice for process monitoring covers that selection, and the same infrared cameras category stocks LWIR and MWIR models.

Two borderline cases deserve a test before a purchase order. When two materials' absorption bands overlap everywhere in the band, the escalation is more discrete bands, not more pixels; hyperspectral imaging: how it works and when it beats RGB and filters explains when that pays. When the contrast you need is only transmission through a wall — fill level through dark HDPE, for example — 850-950 nm NIR hardware does the job without InGaAs at all.

Checks before you order

Spectral proof

• Image a real sample through the candidate bandpass before committing; measured contrast between good and bad units is the only evidence that counts.
• Try two bands, 1450 nm against 1190 nm for moisture, so a later filter change is a swap rather than a redesign.

Optics

• Confirm the lens wavelength range covers your band and the image circle clears the sensor diagonal.
• Rerun the magnification arithmetic at the real working distance and write down the exposure cap from belt speed.

Camera and environment

• Millisecond exposures with active light: uncooled is normally the right call. Long or passive exposures: cooling.
• Hot cells: plan athermalization now, not after the first focus drift complaint.

Light

• Confirm a source at your band with enough power at the working distance, and shield it for eye safety before first light.

Send the sample test result together with the field-of-view numbers: the camera finder, the lens finder and the light finder filter by exactly these values, and the lens step of the configurator builds the chain outward. When one number in the chain refuses to close, contact the applications team — the fix is usually one bandpass away.