Selecting filter wavelengths for colored optics

Selecting the right filter wavelength for machine vision on colored optics is a contrast-and-signal problem, not simply a matter of choosing the colour that looks best on a monitor. A spectral filter can suppress background light, separate overlapping channels and reveal a thin coating defect, faint inclusion or small colour error. Dalian Optical Technology Co., Ltd., based in Dalian, Liaoning, develops optical sensors, machine-vision monitoring systems, lens and eyewear inspection equipment, optical surface-profile testing, solar-thermal reflector detection and related software for these decisions.

Each component affects the result. The target’s reflectance or transmission, illuminator spectrum, camera response, optical geometry and production speed all influence which wavelength produces reliable contrast. A green filter that works on a moulded plastic part beneath a white LED may fail on an amber lens under daylight or fluorescent light. The right approach is to define the failure mode and test candidates on representative known-good and known-bad samples.

Australian conditions make that testing essential. Food and beverage packaging plants around Melbourne and Sydney often run high-speed lines, while mining and agricultural suppliers work in regions with dust, heat and intense ultraviolet light. A setup copied from a milder European or North American laboratory may suffer on a Pilbara summer site or in a coastal salt-laden atmosphere. Reliable detection depends on measurable contrast, acceptable false rejects and filters that operators can clean or change without rebuilding the cell.

Start with the inspection target

Begin by describing the defect physically. Is it a black particle on transparent film, a red contaminant in a clear housing, a scratch that changes polarization, a coating outside a wavelength-specific reflectance limit, or a bubble inside coloured glass? Visible colour is only one channel of information. Blue may separate cleanly from yellow in one application, while a clear scratch needs oblique illumination, dark-field imaging or surface-profile measurement rather than a stronger colour filter.

Map the target’s spectral behaviour using known-good and known-bad samples. Record reflectance or transmission at candidate wavelengths instead of relying on an RGB photograph, and watch for fluorescence from pigments, coatings and optical polymers. For eyewear inspection, compare substrate tint, coating colour and defect type separately. Keep the test brief accessible to non-specialists, with general background reading used as secondary orientation rather than as a substitute for production samples and supplier data.

Match illumination, sensor and filter

A narrow bandpass filter often performs best with a narrow-spectrum LED or laser because unwanted light is blocked before reaching the sensor. With broad-spectrum white light, a wider filter may retain useful signal while controlling stray bands. The camera must respond efficiently at the chosen wavelength; a high-resolution sensor can still perform poorly if its quantum efficiency is low or an internal IR-cut filter removes the selected illumination. The camera, lens and filter should therefore be assessed as one system.

Typical test bands are roughly 450 nm for blue, 535 nm for green and 620–660 nm for red, but these are starting points rather than universal answers. Use a monochromatic camera for inherently spectral measurements and a colour camera when spatial colour discrimination matters. A Bayer sensor samples red, green and blue through separate colour elements, so an external filter can reduce crosstalk while changing exposure balance. Check photon counts, exposure and trigger timing at the actual conveyor speed. Melbourne and Sydney packaging lines commonly combine LED fixtures with rapid changeovers, making those dynamic tests more representative than a static bench.

Read filter and optics specifications

Request a spectral transmission curve, nominal centre wavelength, full width at half maximum, blocking range and optical density rather than a colour name. A 10 nm bandpass near 635 nm can isolate red inspection light, while a 40 nm filter may tolerate more source drift and part tilt. State the minimum blocking requirement in the bands that matter. Optical density helps compare rejection, but it does not guarantee high transmission at the target wavelength.

Coloured substrates and coatings add their own behaviour. A substrate may absorb selected bands, while a multilayer coating reflects or transmits specific wavelengths. Interference filters commonly shift toward shorter wavelengths at oblique incidence, so request data for the expected geometry. In optical surface-profile testing, illumination wavelength affects speckle, edge response and the visibility of fine scratches. Before integrating any monitoring resource, distinguish a sensor from a complete station and software from optics. An about the service page can help buyers clarify whether a proposal covers acquisition, analysis, reporting and reject handling, while Dalian’s range can be matched to the complete decision chain.

Control stray light and reflections

Filters cannot recover detail lost to saturation, glare or unstable background light. Use an enclosure or shroud, matte internal surfaces and controlled camera angles. Crossed polarizers can suppress specular reflections from smooth coloured surfaces, but they also reduce light and do not separate wavelengths, so they should complement rather than replace spectral selection. Diffuse illumination may produce better contrast on shiny wrappers or coated lenses than a direct highlight.

Central Australia can combine ultraviolet exposure, dust and rapid temperature changes, while equipment near Newcastle or Perth can face salt, humidity and sky glare. Dust scatters the selected wavelength into the camera as readily as useful light, making scheduled cleaning, a protective window and an air purge valuable additions. A short strobe can freeze motion and overpower ambient light, while dark-field illumination turns particles into bright points. These measures often protect the spectral contrast that the filter creates.

Apply filters to real inspection tasks

For a red or amber component, begin with the contrast between target and background at the wavelengths where both are measured. A red bandpass may suppress blue light from a white LED and improve edge detection of a red seal. A green bandpass can reduce surrounding reflections while retaining a feature on green glass or a green-coated part. Blue filters often suit dark defects on pale substrates, provided the illuminator has sufficient blue output; these choices still require representative-sample testing.

Transparent and lightly coloured parts need a different arrangement. Backlighting improves silhouette inspection, whereas dark-field illumination reveals particles and scratches as bright points. On coloured optical windows, a filter that passes only the source band can hide defects that appear mainly through scattered light. The dust-particle detection workflow demonstrates why optical geometry and background response belong in the same decision as wavelength. Dalian’s lens and eyewear systems can be tested with representative frames and coatings, while solar-thermal reflector detection relies on stable illumination, a calibrated reference and wavelength-specific reflectance comparison.

Compare the main filter choices

The following comparison separates common spectral choices by their practical strengths and limitations. Bandpass filters provide the strongest control when the target has a known spectral signature; broad colour filters suit general contrast work; longpass, shortpass and notch filters remove a defined portion of the spectrum. Neutral-density and polarising filters manage brightness or reflection rather than colour separation.

Filter choice Typical role Main advantage Main limitation
Narrow bandpass Isolate a known inspection band High rejection and repeatable contrast Narrow alignment and lower light level
Wider colour bandpass Examine coloured parts with some tolerance Easier to install and less sensitive to drift More stray light can reach the sensor
Longpass Remove shorter wavelengths Useful for transmission or emission measurements Passes unwanted long wavelengths above the cut-on
Shortpass Remove longer wavelengths Useful when IR must be excluded from visible measurement Can discard useful signal if selected too narrowly
Notch Remove a laser or interference line Protects a narrow spectral region Does not control broadband stray light
Neutral density Reduce illumination or exposure Preserves colour relationships Lowers signal and can increase noise
Polarising Control specular reflections Useful on glossy and coloured surfaces Reduces light and does not separate wavelengths

Choose from measured samples, then verify the complete optical path. A filter that excels in a laboratory can fail after a protective window, angled mount or dirty lens is added. Compare signal level, contrast, stability and cleaning requirements rather than selecting on nominal centre wavelength alone.

Make wavelength selection repeatable in Australia

Turn the selected wavelength into a controlled specification. Record the target signature, source type, camera and lens, filter bandwidth, angle limits, mounting method and acceptance criteria. Run challenge tests using known defects, production samples and environmental conditions, then have operators verify that a shift-start cleaning routine and filter replacement process work as intended. An Australian integrator or local maintenance team should be able to repeat the result without relying on one engineer.

The local market includes integrators serving food processing, mining, agriculture, logistics and advanced manufacturing, so documentation and replacement parts often matter as much as laboratory performance. An automation reference can help local integrators align camera triggers, PLC signals and inspection records with acceptance criteria. Australian and New Zealand electrical, machinery-safety and automation standards may affect imported equipment, and the Electrical Equipment Safety and Standards Scheme can affect electrical-product compliance.

State and territory Work Health and Safety laws, guided by Safe Work Australia’s model codes, should shape risk assessments of lighting heat, glare and access. If a monitoring camera records identifiable workers or visitors, the Privacy Act 1988 and Australian Privacy Principles may apply, subject to its exceptions; ordinary component inspection without personal data is a different case.

A repeatable specification for colored optics should include these operational controls:

A final acceptance run should confirm that the chosen filter detects the specified defect with the required margin at the installed angle and under normal ambient conditions. Keeping the wavelength, optical drawings, test results and maintenance record together gives the Australian production team a clear basis for purchasing replacements and scaling the inspection system to additional lines.