NIR fluorescence imaging band selection at 800 nm

Fluorescence in the 800 nm near-infrared is the workhorse of indocyanine-green-class tracers: a dye that absorbs red light and re-emits it a little further into the red, marking blood vessels, leaks, coatings or printed security features that a plain camera cannot see. The camera side of such a system succeeds or fails on one decision — which wavelengths each channel captures — because the absorption and emission bands overlap and the emission is faint. This explainer sets the numbers for an 800 nm-class fluorophore: where to excite, where to cut on, which sensor, which lens, and how the channels combine into one fused image, with a worked station layout at the end.

What wavelengths does an 800 nm fluorophore absorb and emit?

An 800 nm-class fluorophore such as indocyanine green (ICG) absorbs across 600-900 nm with the peak near 800 nm once it binds to plasma protein, and it emits fluorescence across roughly 750-950 nm with the intensity peak at 820-830 nm.

Published ICG photometry gives the detail worth designing against: in plain water the absorption peak sits at 780 nm, and binding to protein shifts it to about 805 nm. Perfusion measurements excite at 784 nm and collect the emission peak at 830 nm. Those two windows overlap across most of their width, which is the entire difficulty of this imaging — the light you shine in and the light you collect share territory.

Machine-vision stations meet the same chemistry outside the clinic. ICG-class cyanine dyes serve as leak tracers in fluid circuits, as coating markers and as near-infrared security inks, and the spectral numbers carry over unchanged. The station design therefore fixes two numbers before any hardware: excite at 780-785 nm, and detect from about 820 nm upward. If the question is still which band the inspection needs at all, start with SWIR vs NIR vs visible imaging: which band to inspect with.

How far apart are the excitation and emission bands, really?

Only 35-45 nm: exciting at 785 nm against an emission peak at 820-830 nm leaves that narrow gap, so the bands overlap heavily and a filter edge — not geometry — separates the fluorescence from scattered excitation light.

That gap is why the emission filter carries the system. A hard-coated long-pass filter moves from blocking to passing within a few nanometres to about 10 nm, so a cut-on at 820 nm passes the 820-900 nm emission and blocks 785 nm scatter at the same time. Order blocking of OD 6 at the excitation line, the level standard fluorescence filter sets specify; leakage at OD 2 or OD 3 sits far above the fluorescence signal and swamps it. Scatter and ghost reflections inside the housing add their own background — machine vision stray light: diagnose low image contrast covers that hunt.

The excitation side — source choice, dose and the clean-up filter in front of it — has its own numbers, and fluorescence inspection lighting: excitation and filters works through them.

Can a color camera see 810-900 nm fluorescence?

Not as it ships: a color machine-vision camera carries an IR-cut filter blocking roughly everything above 650 nm, and its Bayer mosaic spends sensor area on colour reconstruction that a weak NIR signal does not need.

Some cameras make the IR-cut filter removable or switchable, and color camera near infrared imaging: IR-cut filter options shows what changes when it comes out — the red channel picks up most of the NIR, and the visible image gains an infrared component. That is useful for a feasibility test and wrong for a production detection channel. Commit to this split:

• Detection channel: mono. Every pixel collects emission, calibration is one curve, and the faint signal gets the whole sensor.
• Reference channel: color, IR-cut in place. The visible image stays colour-correct and free of NIR contamination, which keeps the overlay meaningful.

For the reference channel, a color global-shutter camera holds the visible image sharp on moving parts; the MV-CU020-90GC (2.5 MP, 1/1.7" sensor, GigE, global shutter) is a sound pick.

How do you capture the visible and NIR images at the same time?

With a dichroic splitter and two cameras, one sensor per band: that is the default for machine-vision fluorescence stations because it sends nearly all of each band's light to its own sensor and has no moving parts.

Prism multisensor cameras are what fluorescence endoscopy devices use: two or three sensors behind one objective capture visible and NIR in the same instant, and the NIR channel is overlaid on the visible image as false colour. When a task needs many narrow bands rather than two wide ones, hyperspectral imaging: how it works and when it beats RGB and filters is the next step.

For the overlay itself, render the NIR channel as green on a grey visible image, and trigger both cameras from one hardware pulse so the pair matches in time — synchronize two cameras for ROI image comparison and overlay covers the wiring.

Is silicon still sensitive at 810-900 nm, or do you need SWIR?

Silicon: a typical front-illuminated CMOS sensor still converts 20-40% of photons at 820 nm, so for ICG-class dyes a mono silicon camera is the right detector and SWIR adds nothing at the emission peak.

The conversion falls toward the long end, reaching about 10% by 900 nm, so design for 20% at the 830 nm peak and treat the 900-950 nm tail as a bonus. Standard InGaAs SWIR sensors respond from roughly 900 nm to 1700 nm: their response has barely begun where ICG peaks, so a SWIR camera is the wrong spend for this dye. Earn it only when the label itself emits beyond 1000 nm, or when the same station inspects in the short-wave infrared for another reason.

The MV-CI013-GS-NN is a 1.3 MP SWIR camera with the IMX990 sensor, GigE interface and a fanless housing.

For coarser imaging past the silicon cutoff, browse infrared cameras. The LS-IRC-0029 is a 0.3 MP SWIR model with 15 μm × 15 μm pixels and a global shutter. The emission channel itself belongs in area scan cameras.

What lens and emission filter go in front of the sensor?

A lens specified into the near or short-wave infrared, plus a long-pass emission filter with its cut-on at 820 nm and OD 6 blocking at 785 nm, mounted as close to the sensor as the C-mount thread allows.

Lens transmission is the quiet trap. A lens corrected and coated for visible only can lose transmission in the 850-950 nm range, exactly where the emission lives, and its coatings are not specified there. SWIR-rated glass is specified across the whole band: the LS-SWL-0246 is a 25 mm, F1.8 C-mount SWIR lens rated 400-1700 nm. Its image circle covers up to 8 mm, so pair it with 1/2" and smaller formats.

Three more rules for the detection path:

• Run the detection channel wide open. Emission photons are the scarce resource; an F1.8 aperture passes about five times the light of F4 ((4/1.8)² ≈ 4.9).
• Focus at the working wavelength. A lens focused in white light sits slightly out of focus at 830 nm, so focus the detection channel through its emission filter with the target lit by the excitation source.
• Match the image circle. The sensor diagonal must sit inside the lens's stated circle or the corners go dark; SWIR lens and UV lens basics: what changes outside visible light gives the numbers to check.

How do you size the camera and lens for a fluorescence station?

Work from the field of view backward: magnification is sensor width over field of view, and focal length and working distance follow — worked here for a 50 mm tray, a 1/1.7" sensor, a 25 mm lens and a 190 mm working distance.

Take a 50 mm wide tray to screen for tracer deposits, with the lens fixed at 25 mm to keep the station compact. A 1/1.7" sensor has an active area of about 7.6 mm × 5.7 mm and a diagonal of about 9.5 mm. The LS-CMR-0166 puts 4000 × 3000 pixels on that format at 30 fps over USB3.

The LS-SWL-0207 is a 25 mm C-mount SWIR lens with an 8-11 mm image circle, so it covers that diagonal. Across 4000 pixels, 7.6 mm is 1.9 μm per pixel on the sensor.

Magnification is m = 7.6 / 50 = 0.152, so the working distance is WD = 25 × 1.152 / 0.152 = 189 mm — mount the camera about 190 mm above the tray. The field then covers 50 mm × 37.5 mm (5.7 / 0.152 = 37.5 mm), and the pixel scale is 50 mm ÷ 4000 pixels = 12.5 μm per pixel on the part, so the smallest feature worth trusting is about two pixels, 25 μm. When the defect size is the known quantity instead, run it the other way with the camera resolution calculator.

Exposure: at 30 fps the frame period is 33 ms, so set a 20 ms exposure and gate the excitation pulse to the same window. If the tray moves at 5 mm/s, 20 ms smears the image by 0.1 mm — eight pixels at 12.5 μm each. On a moving line, drop to 5 ms (25 μm of smear, two pixels), pulse the excitation harder rather than raising gain, and prefer a global-shutter model so the rolling scan does not bend geometry. Depth of field at F1.8 is about 0.7 mm: 2 × 1.8 × 0.0038 × 1.152 / 0.152² = 0.68 mm, using a two-pixel circle of confusion (2 × 1.9 μm = 3.8 μm). Stopping to F4 buys about 1.5 mm at the price of light.

When is an 820 nm cut-on the wrong choice?

When the dye's environment shifts its spectra: binding moves the absorption peak from 780 nm in water to about 805 nm on protein, high concentrations quench ICG fluorescence instead of brightening it, and a label emitting beyond 1000 nm calls for a SWIR camera.

Three cases change the filter decision:

• Bound or concentrated dye. The emission shifts with the same binding that moves absorption, and aggregated ICG quenches. Measure the dye as it will sit in the process, then place the cut-on 15-20 nm above the excitation line.
• Thick or wet samples. Water dominates absorption in tissue-like media and scattering falls as wavelength rises, so collected emission skews long and the long-wavelength tail earns its keep. Near infrared wavelength selection for water and silicon inspection covers those water windows.
• Strong NIR ambient. Daylight and hot lamps carry energy at 830-900 nm. Swap the wide long-pass for an 830/40 bandpass, or pulse the excitation and gate a short exposure against it.

For labels that fluoresce under ultraviolet instead — coatings, adhesives, clear plastics — the camera side changes completely, and reflected UV vs UV fluorescence imaging: which one you need sorts that choice.

What should you check before ordering?

Seven checks: excitation centre, emission cut-on and blocking, a mono detection sensor with no IR-cut in its path, a lens rated past 950 nm, one trigger shared by both channels, and a dye-on-paper bench test before the layout is committed.

1. Excitation centred at 780-785 nm. Keeps scattered light 35 nm below the cut-on, where a hard coating blocks at OD 6.
2. Emission filter cut-on at 820 nm, blocking OD 6 at 785 nm. The pair holds the 820-900 nm band and rejects the excitation.
3. Detection camera mono, IR-cut free. Every pixel then collects emission, and one flat-field calibrates the channel.
4. Sensor conversion of 20% or more at 830 nm. Front-illuminated silicon meets this as a rule; design for 20%, not the brochure peak.
5. Lens transmission specified through 950 nm. Visible-only coatings can sag inside the emission band.
6. One hardware trigger for both cameras. Fusion without parallax or timing slips between channels.
7. Bench test with dye on white paper. At the planned working distance, exposure and excitation power; confirm the overlay before the frame is built.

Then size the optics with the calculator above, work through the camera finder and the lens finder, or start the configurator at the camera step. If the dye is unusual, send its excitation and emission curves through the contact page and we will check the filter edge against them.