SWIR Camera Lens for Fruit and Vegetable Sorting: Bruise Detection, Internal Quality, Surface Defects and Automated Grading

Automated fruit and vegetable grading becomes considerably more difficult when product quality cannot be judged reliably from colour, size and surface appearance alone. A fruit may have acceptable colour while containing early-stage bruising beneath the skin; two vegetables with similar external appearance may differ in internal condition; and mechanical damage, chilling injury, tissue deterioration or early decay may begin before a clearly visible defect develops. This is where a SWIR camera lens for fruit and vegetable sorting can support a more informative inspection architecture by allowing the imaging system to work with wavelength-dependent differences associated with tissue structure, water distribution and material composition rather than relying only on visible appearance. Modern hyperspectral research confirms that spectral imaging can provide both spatial and spectral information for evaluating internal and external quality deterioration in fruits and vegetables, including mechanical damage, surface defects, browning, wilting and other quality changes.

For an OEM, however, the objective should not be simply to add SWIR imaging to an existing grader. A production sorting machine must answer much more specific questions: What defect must be detected? How early must bruising be found? Is the target condition on the surface, immediately below the skin or deeper inside the product? What is the smallest damaged region that can change the grade? How quickly is the produce moving? Can every side of a curved fruit be inspected? How much product variation exists between cultivars, seasons and maturity stages? These questions determine the optical geometry and ultimately whether an 8.5 mm, 12.5 mm, 25 mm, 35 mm or 50 mm SWIR lens is appropriate.

The Kyptec Automation® SWIR Camera Lens collection provides five dedicated focal lengths within a common 900–1700 nm, 2 MP, 2/3-inch, F1.4 and C-Mount platform. Kyptec Automation® also identifies food and beverage processing among the intended application areas of this SWIR lens family. This gives fruit- and vegetable-equipment OEMs a practical way to select different inspection geometries while remaining within one dedicated SWIR optical category.

Fruit and Vegetable Grading Needs More Than Colour Inspection

Visible cameras are highly effective for colour grading, shape measurement and obvious surface blemishes. Their limitation appears when quality changes are spectrally or structurally different without producing a strong visible signature. A bruise may begin as damaged subsurface tissue before skin colour changes significantly. Internal deterioration may develop beneath an intact exterior. Differences in soluble solids, moisture distribution, texture or maturity can also influence spectral response even when two products look similar.

Hyperspectral imaging is useful in this context because it combines spatial information with spectral information. Reviews of fruit and vegetable inspection describe applications covering internal quality, external defects, maturity, biochemical characteristics and quality deterioration, while also highlighting the importance of handling non-flat product geometry correctly.

A SWIR lens is therefore not being selected simply to “see infrared.” Its role is to form a stable image of the produce across the spectral bands used for classification while preserving enough object detail for the grading algorithm to identify where the abnormal region occurs.

Bruise Detection Is Fundamentally a Tissue-Contrast Problem

A bruise is more than a dark mark. Mechanical impact damages cellular structure, alters scattering inside the tissue and can change local water distribution and biochemical state. These changes can modify the way SWIR radiation interacts with the damaged region before the bruise becomes strongly visible.

The engineering implication is important: bruise detection should be developed around measurable separation between healthy and damaged tissue, not around whether the bruise “looks dark” in one image.

A robust development program should include freshly bruised samples, partially developed bruises, fully visible bruises and healthy controls. If the intended machine must identify damage soon after impact, testing only mature bruises will create an unrealistically easy problem.

The system should also establish how bruise contrast changes with time after impact because the spectral signature of damaged tissue may evolve during storage and handling.

Internal Quality and Subsurface Quality Are Not the Same Thing

The phrase “internal defect detection” is often used too broadly.

A defect a few millimetres below the skin is a different optical problem from a condition deep inside a large fruit. Light penetrating biological tissue is gradually attenuated by absorption and scattering. Thick rind, skin structure and internal composition can limit how deeply useful information can be recovered. Reviews of nondestructive fruit inspection specifically note that penetration capability varies between products and that thick-skinned fruits can be more difficult to assess internally.

An OEM should therefore specify the approximate defect depth.

The useful question is not:

Can SWIR see inside fruit?

It is:

Can the selected wavelength and optical geometry produce repeatable contrast from the required defect at its maximum production depth?

That distinction prevents unrealistic system requirements.

Surface Defects Still Matter in a SWIR Grading Machine

A SWIR sorter does not have to ignore external defects simply because the technology can reveal information beyond visible appearance. Surface damage, decay, bruising, scars and abnormal tissue can all alter the SWIR response.

The difficulty is that not every surface variation should change the commercial grade.

Natural skin texture, stems, minor harmless marks and normal cultivar variations can create spectral and spatial differences. The grading model must therefore distinguish economically meaningful defects from acceptable variation.

This is one reason the training set should contain more good-product diversity than many laboratory demonstrations use.

If only perfect produce is labelled “good,” the machine can become excellent at rejecting normal commercial fruit.

Curved Produce Creates a Major Optical Challenge

A flat calibration target reflects illumination predictably. A mango, apple, tomato or potato does not.

The centre of a curved product may face the camera directly while its edges tilt sharply away from the illumination. Identical tissue can therefore appear to have different intensity depending on surface orientation. Shadows and localized highlights can become stronger than the defect signal itself.

Research on hyperspectral fruit and vegetable inspection identifies morphological correction for non-flat products as an important part of reliable analysis.

For an OEM, this means illumination geometry must be developed before classification thresholds are frozen. Diffuse illumination, multiple source directions and normalization can help reduce geometry-related variation, but the correction strategy should be validated across the real range of fruit sizes and shapes.

One Camera View May Not Inspect the Entire Fruit

A camera positioned above a conveyor cannot inspect a surface it cannot see.

This becomes critical for approximately spherical or irregular produce. A defect on the underside may remain completely outside the optical field even if the SWIR technology could detect it perfectly when visible.

Automated grading machines therefore need to consider product presentation together with spectral imaging. Rollers, rotation mechanisms, multiple views or sequential imaging positions may be required if near-complete surface coverage is part of the grading specification.

The lens cannot solve occlusion.

This is an important purchase decision because an OEM should define whether the SWIR station is intended to inspect one exposed surface, several controlled orientations or essentially the complete product.

Field of View Should Be Calculated From the Grading Lane

The required inspection width should include every valid product position, not simply nominal produce diameter.

Consider a grading lane 300 mm wide carrying three pieces of fruit with some lateral motion. If the camera sees only the theoretical product centres, edge-position fruit can become partially cropped.

The required FOV must therefore include lane width, product-size variation and tracking margin.

At the same time, excessive FOV wastes pixels.

If a 1600-pixel horizontal image covers 400 mm, nominal sampling is 0.25 mm/pixel. A bruise 5 mm wide occupies around 20 pixels before optical blur and curvature are considered. If the same sensor is stretched across 800 mm, the same bruise occupies only about 10 pixels.

That difference can materially change grading reliability.

The Smallest Grade-Changing Defect Should Determine Spatial Sampling

Fruit graders often have formal or commercial tolerance limits. A tiny blemish may not matter, while a bruise above a certain diameter may cause downgrading.

The optical specification should therefore include the smallest defect that changes the grading decision.

This provides an objective basis for deciding how much FOV the camera can cover.

The goal is not maximum optical magnification. The goal is enough pixels across the smallest economically meaningful defect while still covering the required lane.

That produces a much stronger lens-selection specification than requesting “high-resolution fruit inspection.”

Kyptec Automation® KL-1408 for Broad Produce-Sorting Geometry

Where a grading machine needs wide lane coverage from a relatively compact camera position, the Kyptec Automation® KL-1408 8.5 MM SWIR Camera Lens provides the shortest focal length in the current portfolio. The Kyptec Automation® SWIR family is designed for 900–1700 nm operation and the 8.5 mm model belongs to the same 2 MP, 2/3-inch, F1.4 and C-Mount architecture used across the category.

This geometry is useful to evaluate for large produce, multiple objects across a belt or compact sorting-machine designs where camera height is restricted.

Its limitation is the same one faced by every wide-field system: object sampling decreases as physical coverage increases. The minimum bruise or surface-defect size should therefore be checked before committing to the widest available focal length.

Kyptec Automation® KL-1410 for Wide Sorting With Better Sensor Utilization

The Kyptec Automation® KL-1410 12.5 MM SWIR Camera Lens provides a useful step toward a tighter field. Its published specifications include 12.5 mm focal length, 900–1700 nm wavelength range, F1.4 aperture, 2 MP resolution, 2/3-inch sensor format and C-Mount.

For fruit and vegetable grading, 12.5 mm can be valuable when an 8.5 mm view includes excessive empty conveyor or surrounding machine structure.

Reducing unnecessary background means more of the available sensor is used for produce, which can improve segmentation of smaller defects without moving to a substantially longer focal length.

Product Height Variation Changes Both Magnification and Focus

A sorting machine may handle fruit of different diameters on the same conveyor. The top surface of a large fruit is closer to the lens than the surface of a smaller one.

That difference changes both effective working distance and focus condition.

If the depth of field is too shallow, smaller or larger produce can become less sharp even though the nominal grading sample was perfectly focused.

The production aperture should therefore be selected using the full expected height range. The F1.4 maximum aperture across the Kyptec Automation® SWIR range provides strong light-collection capability, but fruit grading may require the lens to be stopped down to increase useful depth of field.

The correct production aperture is the one that preserves both exposure and defect visibility throughout the grading envelope.

Conveyor Speed Must Be Converted Into Motion During Exposure

High-speed grading puts a practical limit on exposure.

If produce travels at 1.8 m/s and exposure is 500 µs, it moves:

1800 mm/s × 0.0005 s = 0.9 mm

during a single exposure.

If object sampling is 0.3 mm/pixel, the motion corresponds to approximately three pixels. A small bruise edge can therefore become noticeably blurred.

Reducing exposure to 150 µs reduces movement to approximately 0.27 mm.

This is where the optical throughput available from an F1.4 SWIR lens can become important. More available light can support shorter exposure, provided illumination and depth-of-field requirements remain compatible.

For buyers, conveyor speed should therefore be included in lens and illumination qualification rather than treated only as a mechanical specification.

Internal Quality Should Be Calibrated Against Real Reference Measurements

If the machine is intended to estimate maturity, soluble solids, firmness or another internal quality attribute, the spectral model needs trustworthy reference data.

Visual labels such as “ripe” or “good” are not sufficiently precise for quantitative development.

Studies of nondestructive produce inspection use reference measurements to relate spectral data to internal quality parameters such as soluble solids, moisture and other compositional variables.

For OEM development, the correct procedure is to image the product first, measure the target quality independently, and then use those paired data to build and validate the prediction model.

The model should ultimately be tested on fruit that was never used during calibration.

Variety and Season Can Shift the Spectral Baseline

Natural products are variable.

Two varieties of the same fruit can have different skin, tissue composition, maturity behaviour and baseline spectral response. The same cultivar grown in different seasons or regions can also vary.

A classifier trained on one homogeneous batch may therefore appear extremely accurate but perform poorly when commercial supply changes.

The production dataset should intentionally include the variability that the machine will encounter.

For multi-variety grading, the OEM may decide either to build one robust model across all approved varieties or to maintain separate product recipes. Both approaches can work, but the decision should be validated rather than assumed.

Maturity Grading Is Better Treated as a Continuum Than a Binary Problem

Many fruits do not change instantly from unripe to ripe.

Maturity is often a continuum involving gradual changes in chemistry, texture, water distribution and external appearance.

A grading machine can therefore benefit from defining several commercially meaningful classes rather than only “good” and “bad.”

Recent reviews describe the growing use of hyperspectral data and advanced classification methods for evaluating maturity and other fruit-quality states, while also emphasizing the importance of model generalization for real-world deployment.

For buyers, this means the optical system should be developed around the required commercial grade boundaries, not merely around a laboratory distinction between two extreme samples.

Defect Detection and Quality Prediction Should Not Be Confused

Bruise detection is primarily a spatial classification task: locate abnormal tissue and determine whether its size or severity changes the grade.

Predicting soluble solids or another internal parameter is different. The machine may use spectral information averaged across a region rather than search for a localized defect.

A single fruit grader can potentially perform both tasks, but their optical and algorithmic requirements differ.

This matters when defining the lens FOV. Local defect detection benefits strongly from spatial sampling, while global composition estimation can sometimes tolerate coarser spatial information if the spectrum remains reliable.

The OEM should state which requirement is dominant.

Kyptec Automation® KL-1412 for More Controlled Quality Analysis

When the inspection zone is narrower and the objective is to concentrate more sensor pixels over individual produce, the Kyptec Automation® KL-1412 25 MM SWIR Camera Lens provides a useful intermediate focal-length option.

This geometry can be evaluated for single-lane grading, controlled fruit presentation or detailed inspection where smaller bruises and localized quality regions need stronger object representation than a broad conveyor field can provide.

The 25 mm class is not automatically more accurate than a shorter lens; it is more appropriate when the machine geometry can use the narrower field without losing valid product coverage.

Longer Focal Lengths Can Support Detailed Secondary Grading

Some production lines separate bulk sorting from detailed quality assessment. A first stage may classify size or remove obvious rejects, after which a second optical station examines individual products more closely.

For this type of tighter inspection geometry, the Kyptec Automation® KL-1414 35 MM SWIR Camera Lens can be a strong option to evaluate. Its published specifications include 35 mm focal length, 900–1700 nm, F1.4, 2 MP, 2/3-inch and C-Mount.

A longer focal length allows a smaller inspection region to occupy more of the sensor at a suitable working distance. This can be useful for premium grading or localized defect analysis where maximum lane width is no longer the primary constraint.

Do Not Train Only on Perfectly Oriented Produce

Laboratory datasets frequently position every fruit similarly. Production graders cannot assume that.

Stem orientation, rotation, tilt and partial contact with neighbouring produce can change both surface visibility and illumination.

The dataset should therefore include deliberately difficult orientations.

If the machine uses rollers to rotate the product, images should be collected throughout the rotation cycle and the algorithm should determine how evidence from multiple views contributes to the final grade.

A bruise visible in only one frame should not be lost because several other views appear normal.

The Grading Decision Should Consider Defect Area, Not Only Defect Presence

A tiny bruise and a bruise covering 20% of the product should not necessarily produce the same grade.

Once the segmentation model identifies a defective region, the machine can calculate defect area, relative defect percentage or distribution across the visible surface.

This supports commercial grading rules more directly than binary detection.

For example, the system might identify that a defect is present but still accept the product into a lower commercial grade rather than reject it completely.

The optical system must provide consistent geometry if area measurements are used, because curved surfaces can cause image area to differ from actual surface area.

Produce Classification Needs an “Unknown” Path

Commercial sorting lines eventually receive products outside the training distribution: an unusual variety, severe disease, attached leaves, packaging fragments or foreign objects.

A classifier forced to choose between only approved fruit grades can assign an unfamiliar sample to the closest known category even when confidence is low.

A stronger architecture includes an unknown or review class.

If spectral or spatial characteristics fall outside the validated range, the product can be routed for secondary inspection.

This protects grading consistency and prevents unusual defects from being silently accepted.

Validation Should Measure Grade Confusion, Not Just Overall Accuracy

An accuracy figure such as 97% can hide commercially important errors.

Suppose nearly all Grade A fruit is identified correctly, but damaged fruit is frequently classified as premium. The overall metric may still look impressive while the machine fails its most important purpose.

OEM validation should examine a confusion matrix across every commercial grade and defect category.

Key questions include: How often is damaged produce upgraded incorrectly? How often is premium produce downgraded? What happens at the exact boundary between Grade A and Grade B? Does accuracy change with fruit position or size?

These questions produce a much more useful acceptance standard than one headline percentage.

A Production Acceptance Test Should Use the Hardest Samples

A strong SWIR grading-machine qualification should include early bruises, low-contrast defects, maximum product size, minimum product size, multiple maturity states, different orientations, expected varieties, edge-of-belt positions and maximum conveyor speed.

The purpose is not to prove that the technology can identify obvious examples.

It is to determine the margin between acceptable and unacceptable products at the commercial decision boundary.

Recent research on automated fruit and vegetable grading continues to emphasize internal and external quality assessment, model generalization and the need to move beyond laboratory conditions toward robust online inspection.

That should also be the OEM's acceptance philosophy.

Why Kyptec Automation® Is a Strong SWIR Lens Platform for Produce-Grading OEMs

The Kyptec Automation® SWIR Camera Lens collection gives fruit- and vegetable-sorting OEMs five focal lengths within one dedicated 900–1700 nm optical platform. Short focal lengths can address broad conveyor and multi-product fields, intermediate focal lengths can support controlled grading lanes, and longer options can concentrate the sensor on smaller inspection regions. The live range uses a common 2 MP, 2/3-inch, F1.4 and C-Mount architecture.

That makes Kyptec Automation® particularly useful to evaluate when an OEM intends to build several grading-machine variants. The optics can be selected around inspection geometry while the broader SWIR platform remains consistent.

Frequently Asked Questions About SWIR Camera Lenses for Fruit and Vegetable Sorting

1. Can SWIR detect a fruit bruise before it becomes clearly visible?

Potentially yes. Mechanical damage changes tissue structure and can alter the spectral response before strong surface discoloration develops. SWIR and hyperspectral imaging research has demonstrated the ability to detect early and subsurface quality deterioration in fruits. Detection timing still depends on fruit type, bruise severity, wavelength and calibration, so the earliest required detection interval should be tested directly.

2. Can SWIR grading identify internal defects through thick fruit skin?

Sometimes, but penetration depth is strongly product-dependent. Thick skin or rind can attenuate useful radiation, and deeper defects may produce insufficient contrast. Products such as thick-skinned tropical fruits require careful optical-geometry and wavelength selection. A successful result on one fruit should not be assumed to apply to another.

3. Can SWIR measure fruit ripeness without cutting the fruit?

Spectral imaging can support nondestructive estimation or classification of maturity-related properties when those properties correlate with measurable spectral changes. However, a predictive model must be calibrated against independently measured maturity or quality references. The lens and camera provide the optical information; the validated model converts that information into a grading decision.

4. Can a SWIR system measure soluble solids or sweetness?

Spectral techniques are widely studied for estimating soluble solids and related internal-quality variables. Practical accuracy depends on fruit type, spectral range, sample variability and the reference method used during calibration. An OEM should not describe the machine as a sweetness meter unless quantitative validation supports that claim.

5. Why can the same bruise appear different depending on where it is on a round fruit?

Surface curvature changes the illumination and viewing angle. The same damaged tissue can therefore produce different intensity at the centre and edge of the fruit. Geometry correction, diffuse lighting and training data containing multiple defect positions are needed to prevent position-dependent grading.

6. Can one overhead SWIR camera inspect the entire fruit surface?

No. A single fixed view cannot see regions hidden on the underside or behind the product. If near-360-degree inspection is required, the machine must rotate the fruit, use multiple imaging views or otherwise expose different surfaces to the camera.

7. Which SWIR focal length is suitable for a wide fruit-sorting conveyor?

Short focal lengths are normally the first candidates when wide coverage is needed from limited camera height. The Kyptec Automation® KL-1408 8.5 MM SWIR Camera Lens offers the broadest geometry in the current family, while the 12.5 mm option can provide better object sampling where slightly less coverage is needed. Exact selection should come from belt width, sensor size and working distance.

8. Is 25 mm better than 12.5 mm for bruise detection?

Not automatically. A 25 mm lens can allocate more pixels to an individual product when the field is narrower, which may help small-defect inspection. A 12.5 mm lens may be more appropriate when several products must be covered simultaneously. The better lens is the one that preserves enough defect pixels while still covering every valid product position.

9. How many pixels should a bruise occupy for reliable detection?

There is no universal number because detectability depends on bruise contrast, optical resolution, motion and the algorithm. However, relying on a defect represented by only one or two pixels is generally fragile. The OEM should establish the minimum reliable pixel footprint experimentally using the smallest grade-changing bruise.

10. Can SWIR detect decay before obvious mould appears?

Some deterioration processes change tissue chemistry, moisture and structure before severe visible symptoms develop, and hyperspectral methods have been investigated for identifying early quality deterioration and microbial-related changes. The exact lead time depends on product and decay mechanism and must be established with verified samples.

11. Why does fruit size affect SWIR grading accuracy?

Fruit size changes surface curvature, working distance, image scale and potentially focus. Larger products may also hide more of their surface from a fixed viewing angle. Training and validation should therefore cover the complete commercial size range rather than only average-sized samples.

12. Should each fruit variety have a separate grading model?

Not necessarily, but variety must be treated as a deliberate validation variable. If varieties have substantially different skin or spectral behaviour, separate recipes may be more reliable. If one model is intended to cover several varieties, all of those varieties should be sufficiently represented during training and independent testing.

13. Can SWIR grade vegetables as well as fruit?

Yes, provided the vegetable-quality condition produces useful spectral or structural contrast. Research reviews cover a broad range of fruits and vegetables and applications including external defects, chemical attributes, quality deterioration and safety assessment. The system must still be calibrated specifically for the vegetable and grading objective.

14. How does conveyor speed affect bruise detection?

Faster movement reduces the available exposure time before motion blur becomes significant. If produce moves several pixels during exposure, small bruise boundaries can become blurred. A large available aperture such as F1.4 can provide useful light-collection margin, but the final exposure also depends on illumination and depth of field.

15. Can a SWIR sorter predict storage quality or shelf-life?

It may be possible to identify spectral features associated with maturity or deterioration that correlate with later storage performance, but shelf-life prediction requires dedicated longitudinal validation. The machine must image known samples and then follow their actual storage behaviour. It should not infer shelf life solely from generic SWIR contrast.

16. Should a fruit grader use reflectance or transmission SWIR imaging?

Reflectance is easier to integrate and useful for surface and some subsurface quality changes. Transmission may provide stronger access to internal conditions when enough SWIR radiation can pass through the product. Thick or highly absorbing produce may make transmission impractical. The decision should be based on the defect depth and the product's optical behaviour.

17. How should an OEM qualify a SWIR grading machine before production?

Use independently labelled samples representing every commercial grade, borderline cases, multiple varieties, product sizes, orientations, conveyor positions and production speeds. Include the smallest required bruise and the earliest required damage stage. Validate the final algorithm on samples excluded from training and measure grade-by-grade confusion rather than only overall accuracy.

18. What specifications should be sent when selecting a SWIR camera lens for fruit sorting?

Provide the sensor dimensions, lane width, required horizontal and vertical FOV, camera-height limits, smallest grade-changing defect, minimum and maximum produce height, conveyor speed, spectral range and whether the product will rotate during inspection. These inputs allow focal length to be selected around the real sorting-machine geometry.

19. Why is Kyptec Automation® a strong option for fruit and vegetable grading equipment?

Kyptec Automation® provides a dedicated SWIR Camera Lens portfolio covering 8.5 mm, 12.5 mm, 25 mm, 35 mm and 50 mm focal lengths within a common 900–1700 nm, 2 MP, 2/3-inch, F1.4 and C-Mount architecture. This allows OEMs to build broad multi-product sorting stations, controlled grading lanes and tighter secondary quality-inspection stations around one specialized optical family rather than forcing every produce application into one field of view.

Conclusion

A SWIR camera lens for fruit and vegetable sorting is most valuable when grading depends on information that visible appearance does not describe reliably. Bruising, subsurface tissue damage, maturity-related changes and internal-quality variation can alter spectral response before the condition produces an obvious visible defect. Research continues to show that hyperspectral and SWIR-capable imaging can support nondestructive evaluation of external defects, internal quality and deterioration in fruits and vegetables, while recent work increasingly focuses on translating these methods toward practical automated grading.

The difficult part is turning that spectral information into a repeatable production decision. A sorting-machine OEM must define the earliest bruise stage that changes grade, the smallest acceptable defect, the maximum product thickness, the required surface coverage, the full size range and the maximum conveyor speed. Curved surfaces, product rotation, illumination variation and seasonal differences must be treated as normal production conditions rather than experimental noise.

The Kyptec Automation® SWIR Camera Lens collection provides a strong optical foundation for solving these different geometries. An 8.5 mm or 12.5 mm focal length can support wider grading lanes where several products must remain visible. The Kyptec Automation® KL-1412 25 MM SWIR Camera Lens can concentrate more sensor area on controlled product streams, while longer focal lengths such as the Kyptec Automation® KL-1414 35 MM SWIR Camera Lens can support tighter secondary grading and localized defect analysis. The common 900–1700 nm, F1.4, 2 MP, 2/3-inch and C-Mount architecture allows these focal-length choices to remain within one focused Kyptec Automation® SWIR optical platform.

For buyers and OEMs, the most reliable selection principle is to design the optical system around the hardest grade-changing condition rather than the easiest sample. If the camera, SWIR lens, spectral bands, illumination, FOV, exposure and product-handling mechanism can repeatedly identify the smallest, earliest and least obvious defect across the full production range, the same system will have a much stronger foundation for automated grading of the easier cases. That is the level at which a purpose-built Kyptec Automation® SWIR Camera Lens becomes genuinely useful for high-quality fruit and vegetable sorting rather than simply adding another imaging wavelength to the machine.