SWIR Camera Lens for Plastic Sorting and Recycling Machines: Polymer Identification, Material Separation and Automated Classification

Plastic recycling becomes difficult when materials that look nearly identical have different polymer chemistry. A clear polyethylene fragment, a polypropylene component and another thermoplastic part may all appear similar to a visible camera, yet mixing incompatible polymers can reduce the quality and commercial value of the recovered material stream. For automated recycling equipment, the optical challenge is therefore not simply to detect whether an object is plastic. The machine must identify material-dependent spectral differences quickly enough to classify individual pieces while they are moving on a conveyor and early enough for the downstream separation mechanism to act on the correct object.

This is where a 900–1700 nm SWIR camera lens for plastic sorting becomes valuable. Polymer molecules contain chemical bonds that interact differently with near-infrared and short-wave infrared radiation. Over the 900–1700 nm region, combinations and overtones associated particularly with C–H groups create spectral structures that vary between polymer families. Research using hyperspectral systems in this range has demonstrated classification of common post-consumer polymers including polypropylene, polyethylene variants, polystyrene, PET and PVC because their spectral signatures are sufficiently different to support automated discrimination.

The lens is an essential part of this measurement chain because it must transmit and form the SWIR image with sufficient contrast, spatial definition and field coverage for those spectral differences to reach the sensor reliably. The dedicated Kyptec Automation® SWIR Camera Lens collection provides 8.5 mm, 12.5 mm, 25 mm, 35 mm and 50 mm focal lengths within a 900–1700 nm, 2 MP, 2/3-inch, F1.4 and C-Mount platform. Kyptec Automation® positions this optical family for material identification and industrial machine-vision applications, which makes it a strong portfolio to evaluate when developing plastic sorting, polymer identification and recycling automation equipment.

Why Different Plastics Can Be Identified in the SWIR Range

A visible imaging system primarily separates objects by properties such as colour, shape, transparency and surface texture. Those features can help identify bottles or packaging types, but they do not directly identify polymer chemistry. Two objects can therefore have almost identical visible appearance while belonging to different resin families.

SWIR classification works differently. Molecular bonds absorb radiation at wavelength-dependent positions. In polymers, C–H, CH₂ and CH₃ related overtone and combination bands contribute strongly to the NIR/SWIR spectral response. Published work on plastic identification notes useful C–H-related features around approximately 1.1–1.25 µm and toward approximately 1.65–1.7 µm, both inside or at the upper region of a 900–1700 nm imaging system.

A sorting machine therefore does not need to infer polymer type from colour alone. It can compare how each object reflects radiation across selected SWIR bands and classify the resulting spectral pattern.

Polymer Identification Is a Pattern Problem, Not a Single-Wavelength Problem

Moisture inspection can sometimes be engineered around a particularly strong absorption region. Plastic identification is usually more complex because the classifier is attempting to distinguish one spectral shape from another.

Imagine that polymer A reflects strongly at one wavelength and weakly at another, while polymer B shows a different relationship. A single grayscale image might not separate them reliably, but several wavelength measurements can create a spectral vector for every object or pixel.

A simplified material signature might be expressed as:

S = [I₁, I₂, I₃ ... Iₙ]

where each (I) is the measured intensity at a different wavelength.

The classification algorithm then evaluates the pattern rather than relying on one brightness threshold. Hyperspectral systems extend this idea by collecting many adjacent spectral channels, producing a three-dimensional dataset containing horizontal position, vertical position and wavelength. Studies in the 900–1700 nm region have successfully used this approach to classify multiple post-consumer thermoplastics.

For an OEM, this has an important consequence: the lens must provide dependable imaging performance across the spectral range used by the classifier, not merely a sharp image at one convenient visible setup wavelength.

A Plastic Sorting Machine Needs Both Spectral Resolution and Spatial Resolution

Knowing what polymer is present is only half of the problem. The machine also needs to know where that polymer is located on the conveyor.

If three plastic fragments pass through the inspection zone simultaneously, the system must assign the correct spectral class to each object and determine its coordinates so the downstream separation device can target the correct piece.

This makes plastic recycling a spatial-spectral problem.

Spectral information answers: What material is this?

Spatial information answers: Where is it, how large is it, and which ejector should remove it?

A SWIR lens therefore has to preserve both material contrast and sufficient object detail. Excessively wide coverage can reduce the number of pixels allocated to each plastic fragment, while an unnecessarily narrow field can leave portions of the conveyor uninspected.

Start Lens Selection With the Conveyor, Not With the Polymer Name

The fact that an application is sorting PET, PE or PP does not determine focal length.

The optical geometry should instead begin with conveyor width, available camera height, smallest plastic fragment, sensor dimensions and expected object-position variation.

Suppose the camera has 1600 horizontal pixels and the system must inspect a 640 mm belt. The nominal object sampling is:

640 mm ÷ 1600 pixels = 0.40 mm/pixel

A 4 mm plastic fragment would therefore span approximately 10 pixels across one dimension before considering optical blur and object orientation.

If the conveyor width increases to 1000 mm with the same sensor, sampling becomes approximately 0.625 mm/pixel, reducing the pixel footprint of smaller fragments substantially.

This calculation tells an OEM far more about practical detectability than the statement “2 megapixel SWIR camera.”

Wide-Area Plastic Sorting: Where 8.5 mm Geometry Becomes Useful

For broad conveyor coverage where the camera cannot be positioned far from the material stream, the Kyptec Automation® KL-1408 8.5 MM SWIR Camera Lens provides the widest focal-length geometry in the current Kyptec Automation® portfolio. The product is specified for 900–1700 nm, 2 MP, 2/3-inch format, F1.4 aperture and C-Mount.

This type of lens can be appropriate for compact recycling machines where broad belt coverage is the dominant constraint. The trade-off is that each fragment occupies fewer sensor pixels than it would under a narrower field. For large bottles, containers or flakes this may be acceptable; for very small particles it may become the limiting factor.

An OEM should therefore choose the 8.5 mm class because the machine requires its coverage, not because the shortest focal length automatically produces the strongest sorting system.

12.5 mm Can Provide a Better Balance Between Belt Width and Particle Sampling

The Kyptec Automation® KL-1410 12.5 MM SWIR Camera Lens reduces the field relative to the 8.5 mm option while retaining the same 900–1700 nm, F1.4, 2 MP, 2/3-inch and C-Mount architecture.

For a recycling-machine builder, this is useful when the shortest lens captures excessive empty belt area or surrounding hardware. Reducing unnecessary FOV means more of the available sensor pixels are assigned to actual material.

That can improve segmentation of smaller plastic pieces and provide cleaner spectral extraction because fewer object pixels are mixed with background pixels.

The important design principle is to use only as much FOV as the conveyor actually requires, plus justified positional margin.

Why Mixed Pixels Can Reduce Polymer Classification Accuracy

Consider a small plastic fragment whose edge falls across a camera pixel. That pixel may contain signal from both the plastic and conveyor background. The measured spectrum is then a mixture rather than a clean representation of either material.

If an object occupies only a handful of pixels, these mixed-edge pixels can represent a significant proportion of the entire object.

This is one reason spatial resolution affects material classification even when spectral resolution is excellent. Research comparing hyperspectral plastic identification at different spatial scales has shown that appropriate spatial sampling is critical as particle size decreases.

For OEM design, the smallest expected recyclable fragment should therefore occupy substantially more than a token number of image pixels.

25 mm Geometry for More Controlled Polymer Analysis

Not every recycling machine inspects an entire wide belt. Some systems analyze narrower streams after mechanical pre-separation, inspect controlled batches, or perform quality control on recovered polymer flakes.

For these applications, the Kyptec Automation® KL-1412 25 MM SWIR Camera Lens provides an intermediate focal-length class that can concentrate more of the sensor on a smaller inspection region.

This geometry is useful where the objective changes from maximum throughput coverage toward material-purity verification, smaller-particle classification or controlled polymer-stream inspection. The same 900–1700 nm spectral compatibility remains available while the field is reduced to improve object representation.

Polymer Sorting Should Be Designed Around Purity, Not Classification Accuracy Alone

A classifier can report high overall accuracy and still produce poor recycling economics.

Suppose a recovered polymer stream requires very high purity, but the classifier confuses a small percentage of another incompatible resin with the target material. Even a numerically impressive overall accuracy may leave too much contamination in the final fraction.

OEM validation should therefore monitor at least three distinct outcomes: recovery rate of the desired polymer, purity of the accepted stream and false-ejection rate of valuable material.

For example, a machine designed to produce a high-purity polypropylene fraction should evaluate how frequently polyethylene or other polymer types enter that stream, not merely calculate total correct predictions across all objects.

Research on polymer recycling specifically emphasizes the importance of polymer-stream purity because the quality of secondary material depends strongly on correct separation.

Labels, Sleeves and Surface Contamination Can Hide the Polymer Spectrum

Real recycling streams are not clean laboratory coupons.

Bottles may carry labels. Containers can contain food residue. Plastic films can be printed. Fragments may be dusty, wet or chemically contaminated. A label can occupy most of the visible surface presented to the camera, meaning the measured spectral signature belongs partly or entirely to the label rather than the polymer underneath.

This is a major reason why training a classifier only on clean virgin samples produces an unrealistic performance estimate.

An OEM dataset should include the actual conditions entering the plant: labels, adhesives, contamination, scratches, ageing, dirt and multiple colours.

The classifier must learn the production population, not an idealized material library.

Why Dark and Carbon-Black Plastics Require Special Caution

One of the most important limitations of conventional reflected NIR/SWIR polymer identification is carbon-black plastic.

Carbon black absorbs strongly across the visible and near-infrared region. The reflected signal can become so weak that the characteristic polymer spectrum is effectively suppressed. This is a well-established limitation of NIR-based plastic sorting and should not be hidden behind claims that all plastic types can be identified equally well.

For an OEM, this means black plastics must be tested as a separate material class.

A 900–1700 nm SWIR system can be very useful for many non-black and suitably reflective polymers, but it should not be specified as a universal solution for conventional carbon-black plastic identification without application testing.

This distinction is technically important and commercially valuable because it prevents the machine from being sold against an optical condition it cannot robustly measure.

Dark Colour and Black Carbon-Filled Plastic Are Not the Same Problem

A dark blue, red or grey polymer may still return enough SWIR radiation for classification because visible colour does not necessarily dictate its entire infrared response.

Carbon-black-filled polymers are different because the pigment itself can absorb broadly in the NIR/SWIR region.

The correct validation procedure is therefore not to group all “dark plastics” together visually. Each relevant colourant and polymer combination should be measured spectrally.

This can reveal which dark materials remain classifiable and which produce insufficient signal.

Thin Films and Transparent Plastics Need Separate Qualification

Very thin polymer films present another challenge. Their optical path through the material is short, so characteristic absorption may be weaker than in thick plastic parts. Transparent materials can also allow substantial interaction with the conveyor background, changing the measured spectrum.

Recent reviews of NIR plastic sorting identify thin films and highly transparent plastics as more difficult cases because spectral contrast can become weak or ambiguous.

The machine builder should therefore include realistic film thicknesses and transparent samples during optical qualification rather than extrapolating results from thick moulded plastic pieces.

Moisture Can Interfere With Polymer Classification

Water itself has strong spectral structure in the SWIR region. A wet plastic object may therefore produce a spectrum different from the same polymer when dry.

In recycling environments where washed material is inspected before complete drying, moisture can become an important nuisance variable.

The classifier should be trained using both expected dry and wet conditions if both can occur in production. Alternatively, the process may need a controlled drying step before spectral sorting.

This is a good example of why polymer classification cannot rely only on a library of ideal spectral signatures.

Illumination Uniformity Is Part of Classification Accuracy

A polymer moving from the center to the edge of the conveyor should not change class simply because it receives less SWIR illumination.

Wide sorting systems therefore need carefully engineered illumination uniformity.

A useful qualification method is to place the same reference plastic at multiple belt positions and compare its normalized spectral signature. If the apparent material response changes substantially with location, the illumination or calibration system needs improvement before machine-learning complexity is increased.

Software should classify chemistry, not compensate for an uncontrolled optical station.

F1.4 Provides Useful Exposure Margin at High Belt Speeds

A fast conveyor creates a limited exposure window. If exposure is too long, moving fragments smear across several pixels, degrading both object segmentation and the spectral signal associated with individual pieces.

The F1.4 maximum aperture specified across the Kyptec Automation® SWIR Camera Lens family provides useful light-collection headroom for short-exposure applications.

That does not mean every recycling machine should operate at F1.4. If particle-height variation creates inadequate depth of field, stopping down may improve focus consistency. The illumination intensity, aperture and exposure should be optimized together at the actual maximum conveyor speed.

Conveyor Speed Must Be Converted Into Image Motion

A practical OEM calculation is to determine how far an object moves during one exposure.

If the conveyor travels at 2 m/s and the exposure is 500 µs:

Object movement = 2000 mm/s × 0.0005 s = 1 mm

If object-side sampling is 0.25 mm/pixel, that corresponds to approximately four pixels of motion during the exposure, which may be unacceptable for small fragment classification.

Reducing exposure to 100 µs reduces movement to 0.2 mm, or less than one pixel at the same sampling.

This calculation links lens aperture directly to sorting performance: stronger optical throughput can support shorter exposure, which can improve spatial localization of fast-moving objects.

Classification Must Be Completed Before the Ejector Decision Window Closes

A recycling system is not complete when the camera identifies polymer type.

The classified object continues moving.

If the distance from inspection line to air-jet or mechanical separator is (D), and belt speed is (V), the available transport time is approximately:

t = D / V

For example, with a 1 m distance and a 2 m/s belt, only about 0.5 seconds are available from optical observation to the object's arrival at the ejector.

Within that time the machine must acquire the data, classify the material, determine its position, predict its arrival and activate the correct separation channel.

Spectral acquisition with unnecessarily large datasets can therefore reduce throughput. Research on hyperspectral plastic sorting specifically identifies computational burden as an industrial implementation challenge.

For OEMs, the best optical system is not necessarily the one that collects the most wavelengths. It is the one that collects enough discriminative information within the required cycle time.

Full Hyperspectral Data Can Be Used to Discover a Smaller Production Band Set

During development, a hyperspectral system can identify which spectral regions contribute most strongly to separating the target polymers.

Once those bands are known, some production architectures may be simplified around fewer wavelengths rather than processing a complete spectral cube for every object.

The exact implementation depends on the plastics and required purity, but the engineering principle is powerful: use broad spectral characterization to discover the discriminative information, then determine whether the final production machine really needs every wavelength.

A 900–1700 nm-compatible optical path provides useful flexibility during this development phase.

35 mm Geometry Can Support Recycling-Stream Quality Control

After initial sorting, recycling plants may need to verify the purity of a narrower recovered material stream rather than inspect an entire mixed conveyor.

A tighter inspection field can be appropriate in this stage.

The Kyptec Automation® KL-1414 35 MM SWIR Camera Lens provides a longer focal-length option within the same 900–1700 nm platform. It can be evaluated for controlled polymer-analysis stations, flake-stream inspection or secondary quality-control systems where greater spatial representation is more important than maximum conveyor width.

This illustrates why Kyptec Automation® provides practical value beyond one sorting geometry: the same SWIR category can support both wide primary inspection and more concentrated downstream quality-control stations.

Training Data Must Represent the Plant, Not the Laboratory

Machine-learning algorithms can classify polymer spectra effectively, and recent research continues to demonstrate successful plastic sorting using 900–1700 nm hyperspectral data and statistical or machine-learning models.

However, no classifier can compensate fully for inadequate training data.

A realistic dataset should include polymer type, colour, thickness, surface finish, contamination, ageing, moisture condition, orientation and the different suppliers or grades expected in the waste stream.

The dataset should also contain difficult negative examples—materials that resemble the target spectrum but should not be accepted.

The final model should then be tested on material batches that were not used for training.

This is the difference between demonstrating polymer identification and qualifying a recycling machine.

Classification Confidence Should Be Used to Protect Stream Purity

An automated sorter does not always need to force every object into a polymer class.

If the spectral response falls outside the validated distribution or classification confidence is low, the machine can route that item to an uncertain or secondary-processing stream rather than contaminating a valuable high-purity fraction.

This reject-unknown strategy is especially useful for unusual additives, heavily contaminated pieces, black polymers and multilayer materials.

For recycling economics, an intelligent “unknown” decision can be better than a confident-looking but incorrect polymer label.

Multilayer Packaging Is Fundamentally More Difficult Than Single-Polymer Parts

A multilayer package can contain several polymers within one object. The surface presented to the camera may therefore not represent the complete structure.

If the top layer dominates the SWIR measurement, the classifier can correctly identify that surface layer while still failing to describe the complete package composition.

OEMs should therefore define whether the machine is intended to identify surface polymer, dominant polymer or overall recyclable category. These are different objectives.

Multilayer structures should be deliberately represented in the validation set rather than treated as ordinary single-resin samples.

Production Acceptance Should Be Based on a Confusion Matrix by Polymer Class

Overall accuracy alone can hide important errors.

An OEM should build a confusion matrix showing how often each real polymer is predicted as every possible class.

If 98% of all objects are correct but a small fraction of PVC is repeatedly classified into a PET stream, that particular error can be far more important than the headline accuracy number suggests.

The acceptance criteria should therefore be tied to downstream recycling consequences.

For each important polymer fraction, define minimum recovery, minimum purity and maximum cross-contamination from critical incompatible materials.

Why Kyptec Automation® Is a Strong Optical Platform for Plastic-Recycling OEMs

The Kyptec Automation® SWIR Camera Lens collection gives machine builders five focal-length options—8.5 mm, 12.5 mm, 25 mm, 35 mm and 50 mm—inside a dedicated 900–1700 nm optical family. The portfolio uses 2 MP, 2/3-inch, F1.4 and C-Mount specifications and is explicitly positioned for material-identification and industrial inspection tasks.

This is particularly useful for recycling-equipment OEMs because polymer-classification machines often need several optical geometries. A broad primary conveyor may need a short focal length, while flake inspection or purity verification may require a tighter field. Kyptec Automation® lets the machine builder solve those geometry changes while remaining within one focused SWIR lens platform.

Frequently Asked Questions About SWIR Camera Lenses for Plastic Sorting and Recycling

1. Can SWIR identify plastic type instead of only detecting that an object is plastic?

Yes. Suitable polymers produce different wavelength-dependent reflectance patterns because their molecular structures create different absorption features. A multi-band or hyperspectral system can use those differences to classify polymer families rather than relying on visible colour or shape. Studies using 900–1700 nm imaging have demonstrated separation of several common post-consumer thermoplastics.

2. Can polyethylene and polypropylene be separated using SWIR imaging?

They can be differentiated under suitable spectral and measurement conditions because their spectral signatures are not identical. NIR/SWIR hyperspectral methods have been studied specifically for PE and PP separation and purity analysis in recycling processes. Industrial performance still depends on sample condition, spectral quality, calibration and the classifier used.

3. Why is 900–1700 nm useful for polymer identification?

This region contains overtone and combination absorption features associated with molecular groups common in polymers, particularly C–H-related structures. Those features create different spectral shapes for different materials and are therefore useful for classification. The range has been extensively investigated for polymer recycling and automated material sorting.

4. Can a SWIR sorting machine identify plastic regardless of colour?

Many colour variations can remain identifiable because polymer classification is based primarily on infrared spectral response rather than visible colour. However, pigments and additives can affect the returned signal. Carbon-black-filled plastics are a particularly important exception because they can absorb so much NIR/SWIR radiation that the underlying polymer spectrum becomes difficult or impossible to recover reliably.

5. Why are carbon-black plastics difficult for conventional SWIR sorting?

Carbon black strongly absorbs radiation across the NIR region, producing very low reflected signal and suppressing characteristic polymer spectral features. This is a physical measurement limitation rather than simply an algorithm problem. A conventional reflected 900–1700 nm system should therefore not be assumed to classify carbon-black plastics reliably without specific testing.

6. Can transparent plastic be identified with SWIR?

Potentially, but transparent and very thin plastics require careful testing. Their short optical path can weaken characteristic absorption features, and background radiation transmitted through the object can influence the recorded spectrum. Optical geometry, conveyor background and material thickness should therefore be included in qualification.

7. Do labels need to be removed before SWIR plastic sorting?

Not necessarily for every machine, but labels can interfere significantly because the camera may measure the label spectrum rather than the underlying container. OEM testing should include realistically labelled objects. Depending on the application, segmentation logic, multiple views or upstream label removal may improve classification reliability.

8. Can dirty or food-contaminated plastics still be classified?

They may be classifiable, but contamination can change the measured spectrum. A production classifier should therefore be trained and validated on the actual level and types of contamination expected in the recycling stream. A system developed only using clean polymer coupons may give misleadingly high laboratory accuracy.

9. Does wet plastic affect SWIR polymer identification?

Yes. Water has strong wavelength-dependent absorption in the SWIR region, so residual moisture can alter the measured spectrum of a plastic surface. If the plant sorts recently washed material, wet samples should be included in classifier development or the drying condition should be standardized before inspection.

10. How small can a plastic fragment be before SWIR classification becomes unreliable?

There is no universal dimension because it depends on FOV, sensor pixel count, lens performance and spectral contrast. The fragment must occupy enough image pixels to obtain a representative spectrum without excessive mixing from the conveyor background. Smaller particles generally require tighter object-side sampling. Research on hyperspectral plastic identification confirms that spatial resolution becomes increasingly important as particle dimensions decrease.

11. Should I use the widest possible lens for a plastic-sorting conveyor?

No. The lens should cover the required belt width plus necessary tracking margin, but unnecessary FOV wastes sensor pixels on empty background. The Kyptec Automation® KL-1408 8.5 MM SWIR Camera Lens is useful where broad coverage is genuinely required, while the 12.5 mm or 25 mm classes can improve object representation when the inspection zone is narrower.

12. How does conveyor speed affect SWIR lens selection?

Higher speed requires shorter exposure to limit motion blur. The lens therefore needs enough SWIR throughput to maintain useful signal during that shorter integration time. The F1.4 capability across the Kyptec Automation® SWIR range provides useful exposure flexibility, although depth of field and illumination still need to be validated under production conditions.

13. What should happen when the classifier is uncertain about a polymer?

For high-purity recycling, it is often safer to route low-confidence objects into an unknown or secondary stream rather than force them into a target polymer class. This prevents unusual, contaminated or multilayer materials from reducing the purity of valuable recovered fractions.

14. Can one SWIR camera identify several polymer types on the same conveyor?

Yes, provided the spectral acquisition and classifier have been trained and validated for those material classes. Research using 900–1700 nm hyperspectral data has demonstrated simultaneous classification of multiple common polymer families. Increasing the number of classes, however, usually requires a more representative dataset and more careful analysis of class-to-class confusion.

15. How should a recycling OEM measure sorter performance?

Measure more than overall accuracy. Important metrics include polymer-specific recall, purity of each recovered stream, false-ejection rate, unknown rate and confusion between critical incompatible materials. These values should be measured using an independent production-like test set containing realistic colours, contamination, shapes and material grades.

16. Can SWIR imaging verify the purity of recycled polymer flakes after sorting?

Yes. A narrower inspection station can analyze a recovered stream and identify spectral outliers representing unwanted polymer contamination. This is a different machine requirement from primary wide-belt sorting and may justify a tighter FOV. A lens such as the Kyptec Automation® KL-1414 35 MM SWIR Camera Lens can be evaluated where a smaller material stream needs greater spatial representation.

17. What data should an OEM provide before selecting a SWIR lens for a plastic-sorting machine?

Provide conveyor width, camera sensor dimensions, available camera height, smallest plastic fragment, maximum belt speed, expected object height, polymer classes, material colours, contamination conditions and whether the system is performing primary sorting or downstream purity control. These parameters determine optical geometry much more reliably than simply asking for a “plastic recycling lens.”

18. Why is Kyptec Automation® a strong choice for SWIR plastic-sorting machine development?

Kyptec Automation® provides a dedicated 900–1700 nm SWIR Camera Lens family spanning 8.5 mm, 12.5 mm, 25 mm, 35 mm and 50 mm focal lengths, with 2 MP, 2/3-inch, F1.4 and C-Mount specifications across the current portfolio. This gives recycling OEMs a practical route from broad conveyor coverage to narrower polymer-quality-control stations while retaining a consistent SWIR optical platform designed for material-identification and industrial machine-vision applications.

Conclusion

A successful SWIR plastic sorting and recycling machine is not simply an infrared camera connected to a polymer classifier. It is a synchronized material-identification system in which molecular spectroscopy, optical imaging, conveyor mechanics, spatial resolution, classification confidence and downstream separation timing must work together.

The 900–1700 nm spectral region is particularly relevant because common polymers produce material-dependent absorption and reflectance structures within this range, allowing many visually similar plastics to be separated according to chemical composition rather than colour alone. Hyperspectral research has demonstrated classification of important post-consumer polymer groups and confirms the value of this wavelength range for automated recycling and polymer quality control.

For machine builders, however, spectral separability is only the beginning. The smallest recyclable fragment must occupy enough pixels for a clean material spectrum; belt illumination must remain stable across the full inspection width; exposure must be short enough to prevent motion blur; labels, moisture, dirt, colourants and multilayer structures must be represented in the validation dataset; and the classifier must complete its decision early enough for the ejector to intercept the correct object. Carbon-black plastics should be treated explicitly as a challenging or potentially unsupported class rather than being hidden inside a broad claim of universal polymer identification.

The Kyptec Automation® SWIR Camera Lens collection provides a strong optical foundation for solving these machine-design requirements. Its 8.5 mm and 12.5 mm options can address wider conveyor geometries, the 25 mm class can support controlled material streams where stronger pixel utilization is required, and longer focal lengths can serve tighter polymer-analysis and recovered-stream quality-control stations. Because the current family maintains a common 900–1700 nm, 2 MP, 2/3-inch, F1.4 and C-Mount architecture, an OEM can adapt inspection geometry without moving away from a focused SWIR lens platform.

For plastic recycling OEMs, the most useful design rule is therefore to qualify polymer chemistry and machine geometry together. First determine which spectral features separate the real production materials; then make sure each object occupies enough pixels, remains sharply imaged at conveyor speed, receives consistent SWIR illumination and can be classified within the available ejector timing window. When those requirements are engineered as one system, a purpose-built Kyptec Automation® SWIR Camera Lens can become an important optical component in reliable polymer identification, automated material separation and higher-purity recycling workflows.