Food packaging inspection lighting: choose by surface and defect

Food packaging hands one vision station every lighting problem at once: a curved printed label, a glossy foil lid, a transparent bottle, a bare aluminum end. A light that flatters the label hides the defect on the lid. This guide gives line engineers and integrators a decision for each station: which geometry to mount, which wavelength to switch to when white light flattens the contrast, and how conveyor speed and field of view set light size and exposure. The numbers come from inverse-square falloff, motion blur and the wavelength filters in the shop's lighting catalog, so you can re-derive each one for your own part before ordering.

Which light geometry matches each food packaging surface?

Three geometries cover almost every food package: a low-angle bar light for print on curved labels, a diffuse dome for shiny containers and caps, and a backlight for edges, fill levels and shape silhouettes. The wavelength, not the geometry, does the foreign-material work.

The rule behind the table is the law of reflection. A glossy surface throws a mirror image of the light source straight into the lens unless light reaches it from every direction, which is why domes and flat diffuse panels win on foil, film and glazed print. A silhouette or a meniscus needs light only from behind — that is the backlight's job. Print needs raking light close to the surface plane, so strokes cast micro-shadows and the ink reads dark against the lit substrate.

The light geometry selection guide for surfaces and features applies the same matching logic beyond food, and the illumination types overview draws each geometry's optical path. If the pack carries a glossy overwrap film, a polarizer on the lens and a polarizing sheet over the light strip the specular sheen: the crossed polarization guide shows the filter layout.

What wavelength shows what white light hides?

Test three wavelengths before buying extra hardware: infrared at 850 nm for foreign material in dry goods, 1450 nm-class infrared for moisture differences, and 365 nm ultraviolet for fluorescent codes on bare aluminum. Each exploits an absorption edge or a fluorescence that white light averages away.

Most packaging inks transmit that infrared band well. Printed patterns that dominate a white-light image fade out, and the structure underneath shows: adhesive splashes, product lumps, and anything in the food that is not food. That single swap finds foreign material in dry goods, powders and pasta where visible light returns a flat gray texture. Where a label crosses the view, the label's own print disappears with it.

Water absorbs strongly around 1450 nm, so moisture differences become gray-level differences: drier and wetter product separates cleanly where visible light shows nothing. Silicon sensors cut off near 1100 nm, so 1450 nm work needs an InGaAs or SWIR camera — a different budget class, justified only when moisture is the decision.

Fluorescent codes run the opposite way: 365 nm ultraviolet excites the ink, which re-emits visible light while the surrounding aluminum stays dark. A band-pass emission filter on the lens blocks the excitation light so only the glow remains.

In the infrared band a monochrome silicon sensor keeps roughly half of its visible-peak sensitivity, so plan about twice the exposure you would use with red light. At 940 nm it falls further still. Color cameras carry an IR-cut filter and see nothing at either wavelength — specify mono for these stations.

The near-infrared lighting guide for printed packaging covers the ink-transmission effect in detail, and the fluorescence excitation and filter guide picks the excitation and emission pair.

For foreign material on curved or glossy product, keep the diffuse geometry and change only the wavelength. One housing can carry all the working bands: the LS-DOM-0001-L4 is a 237.2 mm dome, 629.9 mm long, with three wavelength sets inside, spanning the deep-red and infrared bands covered in this section.

Why does a direct light look uneven on a curved label?

Because LED intensity falls with the square of distance, a close direct array that nails the centre of the label leaves the wrap edges tens of percent dimmer. Keep the stand-off at least twice the lateral extent you must cover and the falloff stays under 20 percent.

An LED at stand-off d0 lighting a point x mm to the side sits sqrt(d0^2 + x^2) mm from it:

At a 100 mm stand-off covering a 50 mm half-width: d_edge = sqrt(100^2 + 50^2) mm = 111.8 mm, so the edge receives (100 / 111.8)^2 = 0.80 of the centre intensity — a 20 percent drop most gray-threshold inspections tolerate. Halve the stand-off to 50 mm and d_edge = 70.7 mm: the edge gets (50 / 70.7)^2 = 0.50, half the centre. That is the saturated middle and gray rim that fails label checks. Decision: stand-off at least 2x the half-width of the area you must cover.

A compact direct ring makes it worse, not better: at close range its output converges at the image centre, so the label middle saturates while the wrap goes dim. Angle a pair of diffused bars 30-45° from the surface plane, one each side of the camera, so the raking light wraps the curve. A 372 x 28 mm white diffused bar covers a full label wrap in one row at that stand-off. The LS-BAR-1276 is that bar: 372 x 28 mm, white, diffused type.

How do line speed and field of view size the light and the exposure?

Size the lit area to the field of view plus about 20 percent margin, then let the conveyor set exposure: at 300 mm/s, holding blur to 0.05 mm takes a 0.17 ms pulse — the figure that decides between constant-on and strobed operation.

Worked example — date code on a bottle label, conveyor at 300 mm/s (use your own line speed; 200-400 mm/s is a typical small-bottle range):

1. Field of view: a 45 x 25 mm label plus 10 mm of position tolerance per edge gives 65 x 45 mm.
2. Smallest feature: a 0.3 mm broken stroke in the date. At 3 pixels across: p = 0.3 / 3 = 0.1 mm/px.
3. Pixels needed: 65 / 0.1 = 650 wide and 45 / 0.1 = 450 tall. Any current industrial camera clears that, so resolution is not the constraint.
4. Blur budget: half a pixel, 0.05 mm. t_max = 0.05 / 300 s = 0.000167 s, about 0.17 ms.
5. Light size: FOV diagonal sqrt(65^2 + 45^2) mm = 79 mm; with a 20 percent margin, order a lit area in the 100 mm class or larger.

Decision: 0.17 ms is strobe territory. Trigger the light from the container-present photoelectric sensor and pulse at rated current, which delivers far more light in that window than the same unit running constant-on. Strobing pays twice — it freezes motion at full pulse current and cuts LED duty cycle, which stretches service life: the L70 LED lifetime guide gives the duty-cycle arithmetic.

On a packaged-goods line the pulse train usually lives in the station sequence rather than the vision code; a worked sequential packaging program for food line controllers shows one structure for that cycle. Sealed trays and pouches read in transmission need the same timing at larger size: the LS-BKL-0865 is a 400 x 300 mm high-power backlight of 92.2 W, the strobe-class panel for that duty.

When is a flat dome or arch light better than a full dome?

Swap the full dome for a flat dome when headroom will not take the 134 mm-plus outer diameter even a compact dome needs, when the camera hole shadows a tall container, or when the pack is long and low — an arch light wraps those in diffusion.

A dome works because its inner wall scatters light onto the part from every direction, so no single specular facet aims the source into the lens. Two things break it on food packaging. The camera hole: on a deep container light arrives steeply, the hole blocks its own cone, and a shadow band sits near the bottom of the image. Height: the dome must clear the part plus its infeed guides.

A flat dome edge-lights a diffusing guide plate and reproduces the dome's even field from a case tens of millimetres thick, so it slides under low guards and lights deep containers to the bottom with no shadow ring. For long, low product — stick packs, biscuit rows, sachet belts — an arch light throws the same diffuse field across the web.

The dome light guide for shiny, curved and wrinkled parts sizes domes and shows where the camera hole starts to bite. Caps, tubs and lids up to about 100 mm across sit with margin inside a Ø116 mm emitting area. The LS-DOM-0085 is a Ø166 mm dome with that Ø116 mm emitting area, sized for single-lane cap and lid stations.

How do you light transparent bottles for fill level and contents?

Light the bottle from behind rather than beside it: a backlight turns the liquid column into a dark silhouette on a bright field, and a 2 MP-class camera fits the meniscus to better than ±0.1 mm — an order of magnitude inside a ±1 mm fill tolerance.

Take a bottle of 70 mm diameter with a 26 mm level window: a 96 x 80 mm field of view covers the window plus skew. On a 1600 x 1200 sensor that is 96 / 1200 = 0.08 mm/px, and a sub-pixel grayscale edge fits the meniscus to about a tenth of a pixel, 0.008 mm. Against a ±1 mm tolerance that is a 100-fold margin — enough to trade down to a modest lens if the budget argues.

Where a wrapper crosses the light path or the liquid contrast runs faint, switch the backlight to the infrared band: many label inks transmit infrared, so the silhouette reads straight through the print.

Fill-level panels fire between triggers rather than pulsing at every bottle, so a driverless constant-on unit keeps the cabinet simple. The LS-BKL-0416 is a 136.4 x 136.4 mm lit-area infrared backlight that runs constant-on with no driver.

The liquid level lighting guide for transparent containers works the meniscus geometry, container shapes and camera angle in full.

What should you check before ordering a food-line light?

Test the wavelength on the bench before anything else, then check the four numbers the order depends on: lit area against field of view, exposure against line speed — 0.17 ms in the worked example — sensor response at the chosen wavelength, and ingress rating against the washdown zone.

• Shoot the part under white and under each candidate wavelength first. White-light contrast often reverses in the infrared band, and the image histogram settles the question in minutes.
• Check lit area of at least 1.2x the FOV diagonal. The margin absorbs container skew and keeps falloff off the field edges.
• Recompute t_max = blur / speed at the real line speed. At 300 mm/s with a 0.05 mm budget that is 0.17 ms; at 150 mm/s the same budget allows 0.33 ms, and constant-on becomes realistic.
• Confirm the camera: mono sensor, no IR-cut filter for either infrared band; above 1100 nm the sensor must be InGaAs.
• Check mounting before size: dome outer diameter plus camera-hole clearance against the guard height, or a flat dome below roughly 134 mm of headroom.
• Rate the light for its zone: anything inside the spray boundary takes a washdown-rated housing, with the driver box mounted outside it.
• Wire the trigger: the container-present sensor into the light controller, pulse width from the exposure number above.

Shortlist by size, wavelength and voltage in the lighting finder, then take the FOV and exposure numbers to the lighting step of the configurator. If the line drawing wants a second pair of eyes, the contact page reaches the applications team, and expert support comes with every order.