(210k) Hyperspectral Imaging Sensors Enhance on-Line Food Quality and Authenticity Inspection.

Used extensively as a Remote Sensing technique in Agriculture and Environmental monitoring, Hyperspectral Imaging (HSI) makes rapid advances in industrial process monitoring and shows great potential to become a valuable technique for rapid qualitative and quantitative assessment of food quality, safety, and authenticity.

Hyperspectral Imaging is a technique that collects spectral information, resolved over 100+ bands, for every pixel in an image. This wealth of spectral and spatial information acquired during a scan can be used to solve challenges like detection of foreign objects or discrimination between types of objects for various applications.

Modern HSI cameras and integrated systems, coupled together with image processing and statistical tools, can be used for product or process characterization and for delivering quantified results.

Significant value can be realized by using HSI as a tool for early inspection at receiving stations in food processing facilities. Here the ability to quickly scan large quantities of product and get results without the need to prepare samples for laboratory analysis can help the user reduce the cost of quality. Better informed decisions can be made early in the process, helping the user assign value to a specific batch or decide that a certain batch does not meet quality criteria. The information from an early inspection can also be used to help improve quality upstream – in the fields, groves and farms.

An example of this type of HSI application is prediction of sweetness or Brix content in oranges. Brix value is one of the most important quality parameters for growers and processors and traditionally requires a sample to be sent to a lab, oranges to be juiced and soluble solids content measured using a refractometer. A case study described in this presentation shows a prediction of Brix based on HSI scan of a few oranges. Refractometer data was used as a reference. Results of the prediction suggest that the technique could be used for quality assessment of samples under controlled conditions. Methodology used to develop a spectral classification model also helped in creating an optimized model for run-time spectral classification. This enabled the system to produce real time Brix results for batches of oranges passing under the scanner.

Another area where HSI delivers significant value is food fraud detection. The technique is well established in forensic applications and art authentication but now, with reduced costs, simplified user interfaces and AI powered algorithms, HSI becomes a usable tool for industrial QC. Food fraud is a major global problem. In particular, products that move through global supply chains and are sold in powder or small flake form are susceptible to foul play. Oregano, for example can be adulterated with cheaper ingredients like myrtle or olive leaf. Detecting this visually is quite difficult and processing facilities often resort to tedious, manual inspection under a microscope or sending samples for a chemical analysis. With HSI the user can quickly scan a sample of the product and develop a model that would differentiate a pure sample from an adulterated one. Further image processing can help in producing quantified results as in % purity, but even a simple color classification can help make a quick decision if the sample is good or if it requires a more detailed analysis.

An important enabler for Hyperspectral technology in industrial applications is the software for real time spectral classification. Machine Learning algorithms allow even a non-specialist user to create practical, deployable models in a matter of hours. Important progress is also being made in sensor design, taking advantage of increasing power of embedded image processing computers complete hyperspectral systems can now be packaged into compact, rugged housings making the technology useful as a process analytical tool. Embedded computing and on-board processing or raw hyperspectral data overcomes one if the major historical hurdles for adoption of HSI – enormous quantities of raw data generated by the cameras. Modern integrated HSI systems with embedded processing can perform run-time spectral classification and produce an output that consists of a waterfall image stream, object metadata or predicted quality attributes like moisture content, Brix and other results of regression modeling.

processing technology in the Hyperspec MV.X is removing the complexity associated with handling vast amounts of raw data. By performing spectral classification and post processing images on board the device we can output actionable results – object locations, size or predicted values directly to the end user.

This solution produces a complete, fully integrated system in a single ingress protected housing. The design lends itself well to being integrated with other automation or operated as a standalone quality monitoring system. Networking capabilities of embedded computing turn it into a smart device that can be deployed as part of a digital solutions strategy. The scalability of having the same system suitable for both small scale bench testing and for full on-line production environment is a big value add for the users.