Fighting Food Waste with Hyperspectral Imaging - Case ImpactVision

Introduction to ImpactVision

ImpactVision is a machine learning company, applying hyperspectral imaging technology to food supply chains in order to improve food quality, generate consistent, high-quality products and reduce waste.

ImpactVision software is aimed at food processors, manufacturers, distributors, and retailers. It provides real-time insights into the quality of foods. For example, the system is able to determine the freshness of fish, the dry matter content of avocados, or the presence of foreign objects – rapidly, non-invasively and at production-grade speeds.

The first product ImpactVision is launching commercially is a foreign object detection (FoD) system that helps food companies who have severe challenges with non-magnetic foreign object contamination. The team are currently working with Mexico’s largest sugar processor, Beta San Miguel. This client has 17 facilities across Mexico, and they want to scale up with ImpactVision across their operations, starting with four warehouses. ImpactVision is currently reaching out to the sugar industry more broadly, alongside other bulk commodities, such as flour, spices and salt, and adjacent sectors like raisins, cookie dough and produce.

ImpactVision’s system combines hyperspectral camera hardware from Specim with their proprietary machine-learning software. The hardware is mounted above a conveyor belt in food processing facilities, in order to capture images of products in real-time, as they pass below. The software models then analyze those images to provide insights about the quality of the product instantly and non-invasively. This information allows food companies to make decisions about how products should be packed, sorted and distributed based on quality attributes such as freshness, shelf-life or the presence of contamination.

Food industry: challenges and the solution

The food industry processing market is huge: In the USA alone, There are 30,000 food processing plants in the USA alone, of which 22,000 are facilities suitable for foreign object detection systems, 5,000 are fruit and vegetable distribution centers and 600 are seafood processors. The top 75 grocery retailers in North America have 533 distribution centers and 50,000 retail stores.

Today, food is inspected with destructive, sample-based methods that only cover about 1-3% of the overall volume, and the need for a more efficient inspection method is clear if the food industry is going to feed a rising global population.

• One-third of all the food produced in the world goes to waste, and 50% of the waste is generated upstream in the supply chain.
• In 2018, there was over 30 high-profile product recalls related to foreign objects.
• Food fraud – adulteration, mislabelling, substitution, unapproved additives and so on – can cost 30-40 billion dollars a year.

With hyperspectral imaging, the inspection coverage can reach 100% of the target food, and it can include measurements that are not yet done at all. It can reliably identify the chemical properties of food products. As a result, the food industry can change from reactive to predictive. This means fewer recalls, less food fraud, and less food waste.

ImpactVision does not strive to become a camera manufacturer: their strategy is to provide the software platform and partner with a hardware provider who will supply the hyperspectral cameras. When selecting a partner, ImpactVision emphasized the following selection criteria for the cameras:

• Good spatial resolution (see as small objects as possible)
• Good spectral resolution (Access many wavelengths in the right part of the spectrum)
• Flexible interface (That the transmission from the camera to the computer is fast enough, could be connected to a Linux system.)
• Affordable price (the total price cannot be too high when selling the ImpactVision system to the food industry, which operates on slim margins).

Why Specim?

After comparing different hyperspectral camera providers, ImpactVision decided that they got the best value for their money with Specim cameras.

With the FX10-series they got a lot of wavelengths in the right part of the spectrum with good spatial resolution. The information captured by the efficient camera pipeline needs less preprocessing steps which helps to reduce ImpactVision’s application overhead. Specim also provided fast throughput and integration, which enables retrofitting into existing food processing pipeline.

When working with Specim, ImpactVision found that the ordering process is simple and that Specim has always been fast at responding and excellent at reacting to any questions and support requests.

Room for development

HSI cameras typically have limited resolution and higher noise than traditional cameras. Even with the recent significant development in the hyperspectral camera technology, further improvements in these areas could open many additional areas of opportunity.

The challenge with every new application is that you don’t know what you don’t know. So the question is, how can you validate that an application will work before you buy a camera? “My advice to people interested in this technology and Specim’s product is to do your research before the purchase to see if your application can be solved by an HSI-camera,” says Abi Ramanan, CEO and founder of ImpactVision.

What next

With the help of Specim’s camera system and hyperspectral images, great insights can be drawn in areas of food grading and quality.  As more information is collected about what are the key differences in the spectral properties of different food products, that information can be leveraged in grading and quality assessment. This database will be a valuable tool to provide better quality distribution and packaging system.

Longer-term, ImpactVision is working with a partner on the development of a handheld device which can address use cases in other parts of the supply chain, such as in the field, or at the retail and consumer level. The hyperspectral camera provided by Specim sets the premise for mobile and robust image acquisition.