Almost seven years ago, FE featured in this column a new “microtag” technology from TruTag Technologies, which could be used to track and trace high-value food items—those that are often stolen and/or counterfeited. For more on this tagging technology, see “A grain of sand, a whole lot of track and trace,” FE Engineering R&D, June 2014.

Borrowing on TruTag’s hyperspectral scanning technology to locate placed microscopic tags in a food product, its sister company HinaLea Imaging has applied the same technology to provide an intelligent, autonomous and cost-effective imaging solution to track down dangerous pathogens—such as E. coli and Listeria monocytogenes in food and beverage products.

The same technology can also be used to detect production and processing anomalies across a variety of food applications or locate crop variability in agriculture, as well as detect harmful contaminants and pathogenic disease. It can also detect the health of vegetation and provide hydration mapping. All of these detection applications require no contact of the instrumentation to the food or plants.

I spoke with Barry McDonogh, general manager of HinaLea Imaging, to find out how this technology works, how it can be applied to food applications, and what kind of results the technology can deliver.

FE: First, what is HinaLea Imaging’s relationship with TruTag Technologies?

Barry McDonogh: HinaLea Imaging is a business division of and sister business to TruTag Technologies. TruTag initially developed a competence in hyperspectral imaging in 2014 when the technology was utilized for detecting and decoding its microparticles for the purposes of product identification.

In 2018, management determined that the hyperspectral imaging technology TruTag had developed could be used in applications other than detection of our proprietary microparticles, and the decision was made to launch HinaLea Imaging. Today, HinaLea Imaging is focused on providing intelligent imaging solutions for the food processing and quality assurance markets.

FE: Just what is hyperspectral imaging?

McDonogh: Matter has a spectral signature based on how it absorbs and reflects light. The study of this light interaction is called spectroscopy. A spectrometer is an instrument that can be used to measure very fine spectral information of a target. Because spectroscopy only provides information for a single point, it works best where the target material is homogenous.

Hyperspectral imaging builds on the power of spectroscopy by providing both spectral and spatial information related to the target. So rather than getting spectral information about one point on the target, you can get spectral information from every point.

FE: How does hyperspectral imaging work?

McDonogh: There are several discrete steps in hyperspectral imaging.

Data collection: The hyperspectral imager will capture a hyperspectral data cube, which contains the entire reflectance spectra for every pixel in the picture.

Once acquired, researchers can utilize software to analyze the data cube and classify areas of spatial and spectral interest within the target. For example: If we image 10 strawberries, one strawberry has spectral characteristics that do not appear on the other nine strawberries.

Labeling: These differences can be subsequently labeled by linking them with a previous knowledge. In our example, we are informed of the differences associated with an unripe strawberry compared to ripe strawberries and subsequently label that strawberry as unripe.

Training: We image more unripe strawberries and train/optimize algorithms to be able to automatically identify them.

Commercial deployment: The intelligent hyperspectral imaging system that contains both hardware and algorithms is deployed at a customer site. It integrates into the production line and delivers actionable intelligence. The system automatically notifies the line operator when unripe strawberries are detected.

FE: How can it detect bacteria infiltrations in food or a beverage?

McDonogh: Hyperspectral imaging systems can be trained to detect bacterial infiltrations utilizing the same process detailed above. In order to identify the bacterial cell, typically hyperspectral microscopy will be required. However, the presence of certain contaminants, such as aflatoxins in food, can be detected through the spectral response they elicit in the food item.

FE: Is this technology bench-based (for a lab) or handheld, or can it be used to detect bacteria on food as it passes by on a conveyor belt? In other words, what is doable now and into the future?

McDonogh: It depends on what you are looking to identify. We have already proved out the identification of certain toxins in foodstuffs, such as aflatoxins found in grain, and those hyperspectral readers can be deployed on the manufacturing line or at the end of the line at an inspection station.

Foodborne pathogen detection requires hyperspectral microscopy. The technology in its current form is for use in a bench-based system for a laboratory. The ultimate goal is for a self-contained system that would sit online and automatically load and analyze samples without any intervention from a technician.

FE: Can this technology be used on liquids passing through a clear glass tube/pipe?

McDonogh: Yes. Clear glass tubes and pipes can be difficult due to the reflectivity of the glass but can be done with appropriate lighting.

FE: Based on the above, how fast can it return quantifiable results?

McDonogh: For some applications like aflatoxin detection in grain, the response is delivered in real-time or quasi-real time (there may be some lag in the image).

For the detection of foodborne pathogens, the imaging and classification itself takes a few seconds. The challenge is finding the cell in large and sparse sample volumes. It is not the case of finding a needle in a haystack, but rather like finding a needle in several haystacks.

Our goal is to overcome this challenge by utilizing high-speed targeting algorithms. We believe that with this capability in place we could perform analysis in under two minutes.

FE: Does this system require or use built-in artificial intelligence/machine learning tools? If so, how do these tools augment the process?

McDonogh: Absolutely, machine learning is key to the intelligent imaging system.

Normal camera output does not allow for use of machine learning in the way that hyperspectral imaging does.

The spectral information is mapped out spatially for every pixel in the image rather than one point. That spectral information can be interrogated, classified and trained so that system becomes intelligent; it then can perform specific tasks automatically.

The time spent on training algorithms and modeling ensures that the output of the system is actionable intelligence rather than data.

FE: What else could this technology monitor in food samples besides bacteria? Can it check for aflatoxins or mycotoxins? What else?

McDonogh: As mentioned, the system can also test for toxins. We have specifically looked at aflatoxins in grain and peanuts.

The system is also able to process a wide variety of food quality assurance applications including food grading, freshness, ingredient uniformity, foreign object detection and mechanical damage.

FE: Have any food applications been tested so far? What types of applications?

McDonogh: Yes. Hyperspectral imaging is already being used in many of the applications listed above across various food types. A sample of the specific use cases we have looked at include:

• Moisture content of various food items, such as green beans, potatoes and meats
• Fat content of meat
• Detection of salmonella, Listeria, E. Coli, Staphylococcus
• Aflatoxin detection in grain and peanuts
• Seasoning uniformity on potato chips
• Ripeness of various berries/palm fruit
• Mechanical damage to fruit
• Quality grading of various food items such as coffee beans and meats.

FE: Is this system available now?

McDonogh: Yes. These systems are available now. They can be deployed as standalone quality assurance stations or as integrated parts of the production line.

The pathogen detection system is not available commercially today. This requires further development work and regulatory certification. Our target is 2023.

FE: Are you looking for any partners within food and beverage?

McDonogh: Yes. Currently, we are working with the U.S. Department of Agriculture for the development of a pathogen detection system. We are also seeking system integrators and sorting companies to work on development of advanced inline systems.

FE: What is the future of this technology?

McDonogh: Conventional machine vision systems are largely ineffective for the stringent evaluation of food and agricultural products. Hyperspectral imaging provides high resolution spectral and spatial information to enable a comprehensive inspection of foods and other products. It can be extended into the UV or IR spectrum to provide details of chemical and structural composition not discernible in the visible spectrum.

We believe that intelligent imaging systems based on hyperspectral imaging will augment or replace existing machine vision systems.

FE: What didn’t I ask that my food and beverage readers would want to know?

McDonogh: Hyperspectral imaging is not a new technology. It has been around for years. However, its adoption has been curtailed by a number of factors:

• Cost of systems
• Overall size of systems
• The complexity and difficulty of use (you needed a Ph.D. to operate one)

The HinaLea Imaging team has been able to address and improve upon all of these factors. Accordingly, the convergence of reduced costs, improved system portability, and the coupling of the optical technology with sophisticated machine learning to make it easier to use, has led to a marked improvement in commercial and technical feasibility and a spike in market adoption. FE