Hyperspectral and Multispectral Imaging: LED Component Selection Guide

Summary: Multispectral and hyperspectral imaging both read a target across many wavelengths to separate materials that look identical to the eye, and the LED light source is what makes each band usable. Multispectral systems use a handful of discrete LED bands chosen for known absorption features. Hyperspectral systems sample contiguous narrow bands and usually pair a broadband or tunable source with a spectrograph, with LEDs supplementing at key peaks. This guide covers the component selection sequence for an imaging illuminator: choosing the wavelength set by target chemistry across UV (365, 405 nm), visible (450, 520, 630, 660 nm), NIR (780, 850, 940 nm), and SWIR (1050 to 1750 nm); the silicon versus InGaAs sensor split that decides your camera; and array architecture, drive, and radiometric specification. An engineer will finish able to turn an imaging task into a buildable LED specification.

Multispectral vs. hyperspectral, and what it means for the light source

The two approaches differ in how finely they sample the spectrum, and that difference decides whether discrete LEDs can serve as the illuminator at all.

	Multispectral	Hyperspectral
Bands captured	3 to 10 discrete bands	100 to 1000+ contiguous narrow bands
Typical light source	Discrete LED wavelengths, often in an array	Broadband source plus dispersive optics, or a tunable source
Where LEDs fit	The primary illuminator	Supplemental high-power light at chosen peaks
Selection problem	Which bands, which LEDs, how driven	Source uniformity and spectral coverage
Best when	Signatures are known and you classify against them	Signatures are unknown and you need to discover them

Multispectral imaging is the LED-native case. You already know the absorption features that separate your accept and reject classes (water at 1450 nm, a fluorescent dye under 365 nm excitation, chlorophyll contrast near 660 and 800 nm), so you pick discrete LEDs at those bands and read a vector of intensities per pixel. Hyperspectral imaging samples the spectrum so densely that a discrete LED set cannot fill the gaps, so those systems lean on broadband illumination and a spectrometer. LEDs still earn a place in hyperspectral rigs as high-irradiance boosters at the bands where the broadband source runs weak. If your problem is "classify against signatures I already know," you are building a multispectral system and the rest of this guide is your selection path. If it is "find signatures I have not characterized yet," plan for a broadband source and treat LEDs as a supplement.

LED illumination vs. halogen: choosing the illumination source

The incumbent illumination source in most hyperspectral rigs is a quartz-tungsten halogen lamp: cheap, truly broadband, and spectrally smooth from the visible deep into the SWIR, which is exactly what a spectrograph wants. LED-based illumination (marketed as hyperspectral LED lighting when packaged for this use) replaces it in a growing share of systems for four reasons:

• Stability and lifetime. Halogen output drifts as the filament ages and dies in the 1,000 to 2,000 hour range; an LED array holds calibration and runs 20,000+ hours, which matters when a hyperspectral camera is re-referenced against a white tile on a production line.
• Heat on the sample. Halogen dumps most of its light output as unusable radiation and infrared heat; food, biological, and pharmaceutical samples cook under it. LEDs put the light only in the bands the classification uses.
• Strobing and modulation. LEDs pulse in microseconds and can be sequenced band-by-band. Using an LED array and a monochrome machine vision camera, frame-synchronized to the strobe sequence, replaces the filter wheel or spectrograph entirely for known-band work.
• Irradiance where it counts. A multiwavelength LED array concentrates high light output at the exact absorption features the system classifies against, where a broadband lamp spreads its energy across bands the algorithm ignores. The feasibility of using an LED array as the illumination source rises with every band you can name in advance.

Halogen still wins when the spectral signatures are unknown and the system must capture the full contiguous spectrum: exploratory hyperspectral imaging applications, lab spectroscopy, and any workflow where tomorrow's classifier may need bands today's design didn't anticipate. A practical pattern for OEM designers is to prototype with halogen plus a hyperspectral camera to discover the discriminating bands, then productize with a discrete LED array at just those bands, which cuts cost, power, and cycle time in the production system.

Step 1: select the wavelength set by target chemistry

Band selection is the first and most consequential decision. Choose each band where the property you are imaging produces a strong, repeatable contrast, not by what is convenient to source. The bands span the full catalog, from UV through SWIR.

Target property	Band	Why this band
Surface fluorescence, contaminant markers	UV-A 365, 405 nm	Excites fluorophores; the emission you image sits at a longer visible wavelength
Color, print, general surface contrast	Visible 450, 520, 630, 660 nm	Direct color discrimination; narrowband beats broadband white for stable classification
Vegetation health, organic vs. inorganic	NIR 780 to 940 nm	Chlorophyll and cell-structure reflectance; NDVI-style contrast against soil and plastics
Tissue and subsurface (biomedical)	NIR 780 to 940 nm	Penetrates further than visible; hemoglobin and water contrast
Moisture, water content, subsurface bruising	SWIR 1450 nm	Strong water absorption peak
Polymer and hydrocarbon identification	SWIR 1650 nm	C-H bond overtone absorption separates resins and oils
Through-fog, silicon, general SWIR imaging	SWIR 1050 to 1200 nm	Just past the silicon cutoff; reduced scattering

A multispectral illuminator rarely needs the whole table. Two to five well-chosen bands usually separate the classes you care about. Add a band only when it resolves an ambiguity the existing bands leave open. Every extra wavelength adds an LED channel, a drive path, and a synchronization step, so band count is a cost and complexity driver, not a free upgrade. For the SWIR portion of the set, the band-by-band tradeoffs and the seven-step component checklist live in the SWIR LED Lighting Guide, which is the pillar for 1050 to 1750 nm selection.

Step 2: resolve the sensor split before you finalize bands

Your wavelength set decides your camera, and the camera is usually the largest single cost in the system. Two detector materials cover the LED catalog, and the boundary between them sits inside the range spectral imaging often wants to span.

• Silicon (CMOS and CCD) is sensitive from roughly 400 to 1000 nm and is effectively blind past about 1100 nm. It covers every UV, visible, and NIR band in the table with one affordable sensor.
• InGaAs covers roughly 900 to 1700 nm (standard) and beyond 2500 nm (extended), and is the only practical choice for the SWIR bands. InGaAs cameras typically cost 10 to 50x more than equivalent silicon cameras.

The practical consequence: a wavelength set confined to UV, visible, and NIR runs on one silicon camera. The moment a set reaches into SWIR (1450 nm for moisture, 1650 nm for polymers), you need an InGaAs sensor for those bands, which changes the system cost tier and often means two cameras with registered fields of view. Confirm your chosen bands sit inside the responsivity window of the sensor you can afford before committing the band set. A band the camera cannot see is a channel you cannot use.

Step 3: build the multispectral LED array

Once the bands and sensor are fixed, the illuminator is a multi-wavelength array. Two architectures dominate, and the choice follows how the target moves and how clean a spectral separation you need.

• Interleaved array. Multiple LED dies at different wavelengths share one substrate or PCB, and their output mixes spatially across the illumination plane. Suited to moving targets (conveyors, freefall chutes) where time-multiplexed sampling is acceptable.
• Sequential (multiplexed) drive. One optical path, multiple LED packages driven in time sequence, with the camera capturing one frame per wavelength. Gives cleaner band separation at a lower frame rate and needs synchronized drive electronics.

Whichever architecture you choose, the same system-level constraints apply:

• Per-channel current regulation. Different wavelengths have different forward voltages and efficiencies, so a single constant-current source will not give uniform optical output across bands. Regulate each channel.
• Spectral overlap. LED emission has a finite bandwidth (FWHM), and adjacent bands can bleed into each other. Where two bands sit close, add a bandpass filter at the detector rather than relying on the LEDs alone to separate them.
• Optical uniformity. Non-uniform illumination produces inconsistent contrast that corrupts classification. Diffusers, light pipes, or integrating optics homogenize the field before it reaches the target. Uniformity, not peak intensity, sets inspection repeatability.
• Thermal management. Several high-power LEDs in close proximity concentrate heat. Metal-core PCB and active cooling are common above roughly one watt, and junction temperature drives both output and lifetime.
• Camera synchronization. In sequential systems the LED drive pulses must be timed to the camera exposure from a common trigger, so each frame captures exactly one band.

Step 4: specify each LED radiometrically

Classification is only as stable as the light, so specify each emitter by radiometric quantity, not by "brightness." The values to hold on a datasheet or request for quote:

Spec	Symbol / unit	What it drives
Center wavelength	nm	Which absorption feature the band reads
Spectral bandwidth	FWHM, nm	Band separation and filter need
Radiant flux	mW	Total optical power emitted
Radiant intensity	mW/sr	On-axis power into a narrow beam
Irradiance at target	mW/cm²	Detector signal at the working plane
Viewing angle	degrees (half-power)	Field coverage and uniformity
Forward voltage / current	V / mA	Driver design and thermal budget

Back-calculate the required irradiance from the camera sensitivity, exposure time, and working distance, then add margin for optical and window losses. Also specify a center-wavelength bin: emitters carry a wavelength tolerance, and a band that drifts can land off the absorption feature you are classifying against.

Package selection for imaging illuminators

Package choice follows board area, height, optical power, and the heat you need to remove.

• SMD footprints (0603, 0805, larger ceramic) suit board-level integration and compact multi-emitter arrays.
• Low-profile flip-chip minimizes package height where several emitters stack in a tight optical assembly.
• High-power ceramic packages carry multiple chips on copper heat-spreaders for long working distances and high irradiance.
• COB (chip-on-board) mounts many chips on one substrate for high output and uniform area illumination, which is the common choice for inspection illuminators and multispectral arrays.

One optics note that applies across the whole spectral-imaging path: standard acrylic and polycarbonate transmit in the visible but absorb strongly in SWIR, so any window, lens, or diffuser in a SWIR band must use SWIR-rated glass, quartz, or sapphire. Silicone LED encapsulants are generally acceptable inside the package up to about 1700 nm.

Component portfolio

Tech-led distributes Marubeni LEDs across the full spectral-imaging range, from UV-A through 1750 nm SWIR, in surface-mount, low-profile, high-power, and custom array packaging. A multispectral illuminator is typically assembled from discrete emitters at the chosen bands, or as a custom module built to an OEM wavelength set with the drive and optical packaging integrated. For SWIR bands, specifications, datasheets, and sample requests are in the SWIR LED product category.

Frequently asked questions

How do I select LEDs for a multispectral imaging system?

Start from the target chemistry: pick each band where the property you are classifying (color, fluorescence, water, polymer) produces strong, repeatable contrast. Keep the set to the two to five bands that actually separate your classes. Confirm those bands sit inside the responsivity window of a camera you can afford (silicon for UV to NIR, InGaAs for SWIR). Then build the array with per-channel current regulation, homogenizing optics, and camera-synchronized drive, and specify each LED radiometrically by center wavelength, FWHM, and irradiance at the target.

What LEDs are used for hyperspectral imaging?

Hyperspectral systems sample contiguous narrow bands, which a discrete LED set cannot fill, so the primary illuminator is usually a broadband or tunable source feeding a spectrograph. LEDs are used in hyperspectral rigs as supplemental high-irradiance sources at specific absorption peaks where the broadband source is weak. If you can enumerate the bands you need, you are really building a multispectral system, where LEDs are the primary illuminator.

What is the difference between multispectral and hyperspectral imaging for LED component selection?

Multispectral imaging uses a small set of discrete bands, so you select individual LEDs at each band and the LED array is the illuminator. Hyperspectral imaging uses hundreds of contiguous bands, so it needs a broadband source plus a spectrometer, and LEDs only supplement. The practical rule: known signatures and a handful of bands means discrete LEDs; unknown signatures and continuous coverage means broadband plus dispersive optics.

Which wavelengths should I use for multispectral imaging?

Choose by target property. UV-A (365, 405 nm) for fluorescence and contaminant markers; visible (450, 520, 630, 660 nm) for color; NIR (780 to 940 nm) for vegetation, organic-versus-inorganic separation, and tissue; SWIR 1450 nm for moisture and subsurface defects; SWIR 1650 nm for polymers and hydrocarbons; SWIR 1050 to 1200 nm for through-fog and silicon imaging. Combine only the bands that resolve the distinctions your application needs.

Do I need an InGaAs camera for multispectral imaging?

Only if your band set reaches into SWIR. Silicon sensors cover roughly 400 to 1000 nm, which handles every UV, visible, and NIR band on one affordable camera. Any band past about 1100 nm (moisture at 1450 nm, polymers at 1650 nm) requires an InGaAs sensor, which typically costs 10 to 50x more and often means a second, registered camera. Resolve this split before finalizing the band set.

Can discrete LEDs replace a spectrograph in a spectral imaging system?

For multispectral classification against known signatures, yes: discrete LEDs at the chosen bands plus a matched camera replace the spectrograph and cost far less. For hyperspectral discovery, no: contiguous-band coverage needs a broadband source and dispersive optics, and LEDs can only supplement. Match the illuminator to whether your signatures are known or still being discovered.

Can LED lighting replace halogen in a hyperspectral imaging system?

For production systems that classify against known bands, usually yes: hyperspectral lighting built from LEDs holds calibration longer, puts no infrared heat on the sample, and strobes in sync with the camera. Keep halogen when the system must capture a full contiguous spectrum for discovery work, or pair the two: a halogen baseline with LED boosters at the weak ends of the lamp's curve.

Related guides

• SWIR LED Lighting Guide is the pillar for the 1050 to 1750 nm bands, with band-by-band tradeoffs and a full SWIR component selection checklist.

Need help selecting an LED set for a multispectral or hyperspectral imaging system? Contact Tech-led engineering for band recommendations, datasheets, and sample requests.