AI and Food Safety: Today’s Challenges, Tomorrow’s Solutions

Author Christy Draus Head of Marketing

In just the past decade, AI has cemented a presence that’s impossible to ignore across every industry you can imagine. The food service industry is no exception, in which fast food and dining have been quick to adopt AI tools. Some chains plan on using it to study the ordering habits of their patrons to tailor menu recommendations, while others are testing (and adjusting) chatbots in their drive-thru ordering systems. But what about further upstream—does AI have any role in food inspection?

AI vs. Machine Learning in X-Ray Inspection

You may be surprised to learn that modern x-ray food inspection systems already employ AI’s not-so-distant relative, Machine Learning (ML). In fact, today’s Eagle x-ray inspection systems all utilize ML to detect contaminants and determine which food products make it through to the next stage. With ML, a computer is supplied with a library of data on the ideal product to get an idea of what to look for when identifying contaminants.

Some Eagle customers opt for a pre-trained model, which has already been given data on a product that aligns with their offerings. Others may opt to create a new model from their own data. When analyzing an x-ray image, if the algorithm detects something that does not align with what it’s been taught to be ‘correct’, it will flag the unit as a reject.

ML can stand on its own, but it is also a key part of the AI puzzle. Simply put, ML is what allows AI to learn from data in real time. Where ML is taught using static parameters as a baseline, AI will compound every bit of ‘knowledge’ it’s gained to inform each decision it makes. Its database is ever-changing. For that reason, ML and AI that have been trained on the same original dataset may eventually come to different conclusions based on what the AI has learned in the field.

Want to see what today’s tech can do? Send us your product for a complementary evaluation. TEST YOUR PRODUCT >>>

How Reliable Is Today’s AI for Automated Food Safety Monitoring?

When it comes to determining whether a product gets rejected or not, AI presents some unique challenges. AI can only go off of what it already knows, but it learns from every single product. It’s important to make sure that it properly learns right from wrong in order to automate your food safety monitoring as much as possible. For example, say that an AI was taught that a certain image composition is a ‘pass’ in a pass/fail system. However, there can be variations in that image composition between units depending on regular product variations, or different x-ray settings as calibrated by the operator. The AI has to be taught that those differences are okay—otherwise it will produce many false rejects.

While an AI model can mark a certain unit as a reject, there’s no way to extrapolate exactly how it came to that decision. As a result, you can’t simply direct the algorithm to not make the same mistakes again. You can only give it more data on what a passable unit looks like, and monitor its behavior to see how well it catches on. This makes it tricky to eliminate errors from the system.

AI pulls from its vast existing library of data, weighs the unit being inspected against all that data, and makes a decision—just like a human would. So, just like when you put a new employee out on the floor, AI cannot be introduced to the product stream without some kind of human supervision from someone with greater x-ray inspection experience to guide it. At least, not yet.

Supporting Deterministic Algorithms with AI

As previously mentioned, today’s Eagle products can be preloaded with a wide variety of pre-trained ML algorithms for automated food inspection. These models, sometimes called ‘deterministic algorithms’, may not be as adaptive as AI—but while that may sound like a disadvantage or shortcoming, it can actually be a good thing. Unlike AI models, these deterministic algorithms provide x-ray machine operators with greater insights into reject reasoning since they follow a linear, predetermined logic path. In addition, they also give users greater control over ‘accept’ and ‘reject’ product parameters.

In today’s stage of development, AI makes more sense as a support system to deterministic algorithms than as a standalone tool. Think of it as another set of eyes on your product. Although it may not be 100% accurate, there’s a chance it may catch something that the ML model or machine operator didn’t. AI has a more holistic view onto an X-ray image and will base its’ final decision on criteria not always obvious for us humans.  However, it does still require attentive monitoring and a large database to make sure it doesn’t start accepting or rejecting products incorrectly.

Enhancing AI Capabilities with Dual-Energy X-Ray Technology

Eagle’s dual energy technology and advanced imaging software could be promising solutions to eventually help boost the accuracy of AI in x-ray food inspection. By providing an image with higher contrast, more crisp details, and reduced noise from ‘busy’ products, our technology already makes it easier for ML models and machine operators to pinpoint contaminants in food products of all kinds. In the future, this technology may very well play a pivotal part in producing the industry’s most accurate AI food inspection models.

Do You Need AI for Automated Food Safety Monitoring?

While AI has the potential to become a more useful tool in food safety automation with time, in its current state it’s just not quite what customers expect it to be. The ideal AI algorithm would be able to learn in real time without supervision, and learn exactly what it needs to without making mistakes—but the technology isn’t quite there yet.

In order to reach the quality that customers expect, we at Eagle are continuously investing in training and improving the AI models we have in development to ensure that they can actually provide a valuable solution upon release. Many other food inspection companies may claim that they are using AI-powered models, and they very well may be. But what they don’t disclose is how much oversight goes into monitoring these systems to make sure they don’t go off the rails.

At the end of the day, it’s not so much a matter of choosing between AI or deterministic algorithms. It’s a matter of choosing between what works reliably and what doesn’t. AI technology may be exciting in its future possibilities, but for now, don’t feel like you have to rush into adopting the tech. AI’s capabilities in food inspection applications are still being explored to their full potential, with plenty of room to grow. In the meantime, if you’re running an Eagle x-ray inspection system, you should be in pretty good shape.