The Camera That Sees What the Eye Cannot
Conventional machine vision systems for food sorting and inspection capture the spatial information that determines the colour, size, and shape of the food product being inspected. These characteristics allow machine vision to sort by grade, identify foreign material of contrasting colour, and remove items whose size falls outside specification. What conventional machine vision cannot detect is the chemical composition information that distinguishes a visually normal piece of produce or meat whose internal quality is compromised from one whose quality meets the specification. A potato with internal bruising whose skin surface is visually normal, a fish fillet with elevated histamine levels that creates a food safety risk without visible indication, or a grain sample contaminated with mycotoxins whose distribution is invisible to surface imaging are all quality defects that conventional machine vision systems pass without flagging. Hyperspectral imaging extends the measurement capability of optical inspection systems from the visible wavelength range that the human eye and conventional machine vision cameras capture to the near-infrared and short-wave infrared wavelength ranges where the molecular absorption spectra of chemical compounds create the spectral signatures that identify composition as well as appearance.
A hyperspectral camera captures a full spectrum of reflected or transmitted light intensity across hundreds of wavelength bands for each spatial pixel of the image it acquires, creating a three-dimensional data cube whose two spatial dimensions correspond to the image and whose third dimension is the wavelength spectrum at each pixel location. The chemical composition information encoded in the spectral dimension allows hyperspectral imaging systems to identify not just what a food product looks like but what it is made of, enabling the detection of contaminants, defects, and quality parameters whose physical manifestation is chemical rather than visual. The machine learning algorithms that process hyperspectral data cubes in real time, trained on large datasets of hyperspectral images of conforming and defective product, have become the commercial differentiator that distinguishes the most capable hyperspectral food inspection systems from those whose spectral data is not fully exploited by the analysis software that interprets it.
Sorting at Speed: The Processing Line Challenge
The commercial deployment of hyperspectral imaging on food processing lines faces the throughput challenge that all optical inspection systems must address in food manufacturing: the inspection system must keep pace with the production line whose throughput is measured in tonnes per hour rather than items per minute. For produce sorting on high-throughput belt or chute conveyors, the hyperspectral camera must acquire, process, and generate ejection decisions for every item on the belt within the millisecond-scale time window available before the item passes the ejection point. The computational demands of processing hyperspectral data cubes at production line speed have been the technical barrier that has most limited hyperspectral inspection deployment to lower-throughput applications where processing speed requirements are more manageable. The improvement in GPU computing performance and the development of optimised hyperspectral processing algorithms that reduce the computational cost of real-time spectral classification have reduced this barrier substantially over the past five years, extending the range of food processing applications where hyperspectral inspection can operate at commercial throughput without sacrificing the spectral analysis depth that its defect detection capability depends on.
Tomra Food's hyperspectral sorting systems, deployed on potato, fruit, vegetable, and seafood processing lines globally, represent the most commercially mature application of hyperspectral sorting in food processing. Its ability to detect internal defects in potatoes through near-infrared spectral imaging of tuber cross-sections at high throughput, and to identify species adulteration and quality defects in fish fillets through the spectral signatures of protein and fat composition, have created commercial value whose food safety and waste reduction dimensions both contribute to the return on investment that processing companies have used to justify hyperspectral inspection capital investment.
Beyond Sorting: Quality Management and Traceability
The hyperspectral imaging data generated by food processing line inspection creates a commercial value layer beyond the immediate sorting decision at the point of inspection. The spectral data captured from every item passing through the inspection system contains composition information whose aggregation across production batches creates the raw material quality database that supply chain traceability and quality management systems can use for supplier performance assessment, process optimisation, and the root cause investigation of quality incidents. A potato processing operation whose hyperspectral sorting system has captured the internal quality distribution of every tonne of potato it has processed can correlate quality parameters with harvest location, variety, storage duration, and processing conditions in ways that continuous manual sampling cannot approach. The data layer that hyperspectral inspection creates is becoming commercially valuable in its own right as food companies invest in the quality intelligence infrastructure that differentiates their supply chain management from competitors whose quality data is less comprehensive.
Top 10 Companies in Hyperspectral Imaging for Food Inspection Globally
- Tomra Food: World's largest food sorting company whose TOMRA 5B and CC6 sorting systems incorporate hyperspectral imaging for defect detection in potato, produce, and seafood applications; its global installed base, its service network, and its machine learning algorithms trained on the largest food sorting dataset in the industry create the commercial depth that new entrants to the hyperspectral sorting market cannot quickly replicate.
- Key Technology: Food processing equipment company whose VERYX hyperspectral sorting platforms serve the French fry, vegetable, and nut processing markets; its integration of hyperspectral imaging with high-throughput vibratory conveyor systems creates the combined product design and process knowledge that specialised hyperspectral camera companies cannot provide without the food processing equipment context.
- Specim: Finnish hyperspectral camera manufacturer whose push-broom hyperspectral cameras are used in food inspection, quality control, and research applications globally; its camera technology is the spectral imaging hardware that both food sorting OEMs and research users build their hyperspectral applications on, creating the component market position that influences the performance of the entire food hyperspectral inspection industry.
- Unitec: Italian fresh produce sorting equipment company with hyperspectral sorting systems for soft fruit, citrus, and vegetables; its fresh produce market focus and its Italian design and manufacturing heritage create the commercial position in the European fresh produce sorting market that global sorting equipment companies compete against on technology and service.
- AWETA: Dutch fresh produce sorting company with optical and spectral sorting systems for apple, pear, citrus, and mango; its internal quality sorting capability using near-infrared spectroscopy and hyperspectral imaging detects flesh disorders, dry matter content, and sugar content in whole fruit without destructive sampling.
- BaySpec: US hyperspectral imaging company with compact and OEM hyperspectral camera systems for food quality and safety inspection; its miniaturised hyperspectral sensor technology creates the integration possibilities for inline and handheld food inspection applications whose form factor constraints conventional hyperspectral cameras cannot satisfy.
- Buhler Sortex: Swiss processing equipment company whose optical and hyperspectral sorters serve the grain, pulse, and speciality seed markets; its SORTEX H sorting system uses hyperspectral imaging to detect grain contamination including mycotoxin-infected kernels and foreign material that conventional colour sorting cannot identify from spectral composition differences.
- Marel: Icelandic food processing equipment company with optical and spectral inspection systems for fish, poultry, and meat processing; its SensorX X-ray bone detection system and its hyperspectral quality inspection for fish complement each other in the seafood processing application where bone detection and quality grading are both commercially critical inspection functions.
- Perception Robic: Canadian company developing hyperspectral imaging systems for grain quality assessment and foreign material detection in grain handling facilities; its grain industry focus and its integration with grain handling infrastructure create the food safety inspection capability in the bulk grain supply chain that retail-focused food companies are increasingly requiring from their suppliers.
- VideometerLab: Danish company developing multispectral and hyperspectral imaging systems for food quality analysis including seed quality, cereal grain purity, and produce quality assessment; its research and quality laboratory instrument position complements the production line sorting applications of the larger sorting equipment companies and serves the upstream quality analysis market that production line inspection data cannot replace.