Plastics Sorting with SWIR Hyperspectral Cameras

During the upgrade of the plastic recycling industry, issues like mixed materials, difficulty identifying black plastics, and distinguishing polymers with similar appearances have always been pain points on production lines. Short-wave infrared (SWIR) hyperspectral imaging technology, using the 900–1700nm wavelength molecular spectral fingerprints, enables non-destructive and high-speed online identification of various plastics such as PET, PP, PE, PVC, PS, ABS, and PA. This has become a key solution for smart sorting of recycled materials. This article analyzes the principles of SWIR hyperspectral sorting, the challenges of implementing it on production lines, key system selection points, and real project results, providing technical reference for recycling equipment integrators, recycling plants, and research institutions.

I. Existing Sorting Challenges in the Plastic Recycling Industry

Plastic waste is composed of a mix of materials, with different polymers mixed together, making it hard for traditional sorting methods to achieve both efficiency and purity. Manual sorting relies on visual judgment, which is heavily influenced by lighting, material contamination, and worker fatigue, limiting processing speed. It’s almost impossible to distinguish PE and PP fragments that look very similar, and labor costs keep rising.

Density separation (float-sink separation), which relies on differences in material density, fails for some engineering plastics with similar densities. After crushing, fine materials and thin films are poorly sorted, and it also generates a large amount of wastewater.

Ordinary visible-light cameras can only identify colors, not the material’s chemical composition. Regular near-infrared devices see a significant drop in recognition rates when encountering black, carbon-filled plastics or materials covered in dirt or oil. Missing high-risk materials like PVC may lead to the scrapping of the entire batch of recycled material.

Many recycling plants face common problems:

1. Black plastics are hard to identify since carbon black absorbs visible and regular near-infrared signals.

2. PE vs. PP and ABS vs. PS look very similar, making them difficult to distinguish.

3. Surfaces contaminated with oil, labels, or dust interfere with detection.

4. Detection frame rates can’t keep up with high-speed conveyor belts.

5. Multi-layer composite films, aluminum-coated films, and engineering plastics are difficult to identify consistently.

II. How SWIR Short-Wave Infrared Hyperspectral Technology Identifies Plastics

The short-wave infrared (SWIR, 900–1700nm) band can produce unique absorption peaks from the C–H and O–H chemical bonds inside various polymer materials, giving each type of plastic a one-of-a-kind spectral fingerprint.

A hyperspectral line-scan camera captures dozens to over a hundred continuous narrow-band spectra at each point on a conveyor belt while retaining spatial image information, generating a “image–spectrum” data cube. The system compares each pixel’s spectrum and models it with algorithms, meaning it can directly determine whether the material at that point is PET, HDPE, LDPE, PP, PVC, PS, ABS, PA, or PC without contact, then sends classification signals to the air-jet sorting mechanism for automatic sorting.

Compared to ordinary single-point spectrometers, imaging hyperspectral systems can handle fragments, films, and materials of varying sizes, covering the entire conveyor width without missing small debris or impurities. Compared to visible light, SWIR is more resistant to interference from dirt, fading, and dark-colored materials and can process most black plastic samples.

Complete Sorting System Components:

1. SWIR line-scan hyperspectral camera (InGaAs detector)

2. Infrared-specific uniform light source module

3. Conveyor belt synchronous triggering device

4. Spectral analysis software with machine learning classification models (PLS–DA, SVM, etc.)

5. Actuation: air-jet valve array receives classification signals to complete the sorting

III. Key Technology Challenges for Industrial Deployment

Even though SWIR technology is theoretically mature, directly selecting equipment can still face pitfalls. Industrial production lines have strict equipment requirements:

1. High-speed imaging capability: Conveyor belts move fast, so cameras must support high line rates and ROI readout; otherwise, motion blur can cause misidentification.

2. Spectral resolution: To distinguish closely similar materials like HDPE/LDPE, sufficient spectral channels and resolution are needed.

3. Signal-to-noise ratio and stray light control: Dusty and complex lighting in factories can reduce sorting accuracy if stray light is uncontrolled.

4. Black material robustness: Carbon black attenuates signals, requiring highly sensitive detectors.

5. Supporting software capability: It’s not enough to just output raw data; software must support sample labeling, model training, real-time inference, and IO trigger outputs for seamless integration with downstream sorting equipment.

6. Structural reliability: Recycling workshops have heavy dust and strong vibrations, so camera bodies need industrial protection and must handle harsh on-site conditions.

Many devices that perform well in labs see a sharp drop in accuracy when moved to production lines, mostly because they aren’t adapted for industrial sorting scenarios in terms of hardware and algorithms.

IV. FigSpec® SWIR Hyperspectral Camera Plastic Sorting Solution

The FigSpec® FS-19 series short-wave infrared hyperspectral camera is specially developed for industrial online material sorting scenarios, addressing various pain points in plastic recycling and sorting. Its spectral coverage is 900–1700 nm, using an InGaAs detector, and it is suitable for the complex conditions of the recycling industry.

Key parameters for plastic sorting needs:

• Spectral range: 900–1700 nm, fully covering the characteristic fingerprint bands of plastic polymers.

  
  

• Spectral resolution up to 8 nm, able to distinguish subtle spectral features of materials like PE, LDPE, PP.

• Supports ROI band selection, allowing selective reading of key bands for plastic identification, greatly increasing line speed, capable of scanning thousands of lines per second to match high-speed conveyor production lines.

  
  

• Low stray light optical design, stray light <0.5%, maintaining stable spectral data even under dust and on-site lighting interference.

• Compact body with standard C-mount lens, supports multiple lens options for installation on different production line widths.

  
  

• Comes with FigSpec Studio spectral analysis software: supports sample labeling, spectral preprocessing, and machine learning model training; directly outputs classification results and trigger signals, making it easy to integrate into a complete sorting system without developing underlying algorithms from scratch.

Actual project performance: Tested in several recycling factories, sorting common plastic fragments such as PET, PP, PE, PS, ABS, PVC, achieving a comprehensive sorting accuracy of over 95%; can also effectively identify some black carbon-filled plastics, reducing the risk of PVC contamination in recycled materials. The system can be used for new sorting lines or upgraded on existing lines, replacing old optical detection units.

Tip: For extremely difficult-to-identify materials like metallized films and multilayer composite plastics, recognition can be further improved by expanding the sample library and iterating the model.

V. Typical Applications for SWIR Hyperspectral Plastic Sorting

1. Post-consumer packaging waste sorting: crush beverage and daily chemical bottles to distinguish PET, HDPE, PP, remove harmful PVC impurities, and improve flake purity.

2. Industrial plastic scrap recycling: crush home appliance shells and automotive plastic scraps to distinguish engineering plastics like ABS, PS, PC, PA.

3. Black plastic special sorting: handle black home appliance plastics and black pipe fragments, overcoming the limitations of normal NIR devices.

4. Thin film material detection: identify plastic film scraps, distinguishing PE film and PP film.

5. Scientific research and laboratory development: spectral study of new biodegradable plastics and modified polymers, development of sorting algorithm models.

6. Equipment integration and support: provide sorting equipment manufacturers with core optical imaging units, delivering complete detection solutions.

VI. Key Points for Choosing a SWIR Hyperspectral Camera – Things to Confirm Before Purchase

Many buyers focus solely on spectral bands when selecting a camera, overlooking how well it matches the production line. Here are the key selection criteria:

1. Spectral Range: For plastic sorting, first ensure complete coverage of the 900–1700nm SWIR band.

2. Line Rate / Line Frequency: Calculate based on conveyor speed and material width, leaving some margin to avoid motion blur. Prefer models that support ROI functionality, as this can significantly increase actual scanning speed.

3. Spectral Resolution: When dealing with similar polymers, it’s recommended not to exceed 10nm.

4. Software Support: Check if model training, real-time inference, and IO output are provided to avoid spending extra resources developing software after getting the camera.

5. Environmental Adaptability: Ensure the device operates stably in dusty workshops with temperature fluctuations.

6. Test Verification: Whenever possible, perform on-site tests with real factory materials; lab samples don’t guarantee the same results on the production line.

VII. Conclusion

With the push for a circular economy, the market demand for high-purity recycled plastics continues to rise, and downstream manufacturers are increasingly strict about impurity control in recycled materials. Traditional sorting methods can no longer meet the production standards for high-end recycled plastics.

SWIR shortwave infrared hyperspectral imaging combined with AI spectral analysis models is moving from laboratory research to large-scale industrial applications. On the hardware side, high-frame-rate, high SNR, easily integrated hyperspectral devices will become more widespread; on the software side, pre-trained plastic spectral libraries and transfer learning techniques can lower the barrier to building models for industrial use.

SWIR hyperspectral imaging technology addresses the material identification challenge in the plastic recycling industry, turning chemical properties that are hard to see with the naked eye into quantifiable spectral signals. Using imaging hardware suited to the production line along with customized algorithm models can effectively improve the purity of recycled plastic sorting, reduce manual intervention, and achieve high-value plastic recovery. Production line evaluation and sample testing can be carried out together with equipment manufacturers using on-site tests.

FAQ Common Questions

Q1: Can SWIR hyperspectral cameras recognize black plastics?

A: Compared to ordinary near-infrared devices, SWIR has better recognition ability for carbon black-filled black plastics, but signals from fully black high-carbon black materials will still weaken. It needs to be paired with high-sensitivity detectors, and the recognition rate can be improved through model optimization. It cannot achieve 100% perfect recognition of all black materials.

Q2: Can SWIR hyperspectral fully replace float-sink sorting?

A: No, they are complementary. Optical sorting does preliminary material screening, while float-sink is used for further purification. Using both together achieves the best purity and productivity.

Q3: Is developing a complete system difficult?

A: Choosing a camera with a complete SDK and supporting analysis software can greatly reduce the integration workload. The equipment manufacturer mainly handles mechanical and air jet control, while spectral recognition is done by the camera's software.

Q4: Can SWIR hyperspectral recognize biodegradable plastics?

A: Yes, as long as the spectral features of the corresponding biodegradable plastic samples are included in the model, they can be distinguished from conventional plastics.

Q5: Will thin film materials fail to be recognized?

A: Thin plastic films and wrinkled films can cause abnormal spectral reflections, so wrinkles and thin material samples need to be included in the model building to optimize model robustness.