Traditionally, visual inspection has been a very manual role in F&B production, but recent advancements are paving the way for more consistent results and efficient processes.
While some lines are already equipped with neural network-based algorithms for pattern recognition, the latest advances in AI have enabled machine vision to take on far more complex tasks.
However, line managers and boardroom leaders can be hesitant about change. Here we speak to KPM Analytics’ subject matter expert, Yuegang Zhao, on the opportunities and drawbacks of AI-powered vision systems.
What is AI vision?
So what exactly is AI vision? In very simple terms it’s a vision technology that enhances the detection or characterisation of an image (in this case of food products) with the help of artificial intelligence.
But it’s not just for RGB-based imaging systems, it can also be used with hyperspectral and multi-spectral imaging to inspect products beyond visible light. It can also integrate with X-ray imaging to enhance product inspection.
“It’s used for a very broad range of applications from quality inspection and foreign material detection to optimising production,” Zhao told Food Manufacture.
Zhao continued: “AI vision technologies are trained on images of real products to recognise patterns, textures, and variations that are almost impossible to identify manually at today’s production speeds.”
AI technology is deeply trained on specific food products and processes, meaning it has an “outstanding capability” to detect anomalies and contextual clues from the line.
“From determining the bake colour consistency of biscuits and the height or volume of a bread loaf to the breading coverage on a chicken nugget, AI vision inspection technologies remove the inherent subjectivity from quality inspections traditionally managed by human inspectors, while also enabling process monitoring, automation, flow optimisation, and foreign material detection at production speed,” explained Zhao.
How is it being used in food and drink today?
Although process in AI has transformed what is possible within vision systems, we are still in the early stages of understanding just how far this technology can go.
For example, AI can streamline process flow by informing conveyor belt speed adjustments and limiting product jams at critical process points. It can also help reduce injury and joint stress by using robotic arms or rejection mechanisms capable to autonomously moving products to their correct process streams.
“In the case of meat, poultry, and seafood operations, AI can be used in animal welfare applications to inspect animals at intake and ensure they are humanely handled as they enter the processing facility. Then, as the animal is portioned, AI can effectively grade various cuts of the meat to maintain quality standards and limit waste,” added Zhao.
AI inspection can also be used to automate quality inspection, as Zhao elaborated: “Many food processing companies commonly remove product samples from the line for routine quality inspection of certain traits like colour, shape, size, volume, or other measurements.
“Since an AI vision system is carefully trained on the product line and process, inspection can be done on 100% of products on the processing line, including top and bottom inspection.”
AI vision inspection especially excels in foreign material detection, particularly soft foreign materials from product surfaces like rubber, paper, plastic, wood, and other less-dense materials that cannot be identified by X-ray or metal detectors.
“Foreign materials can infiltrate food facilities through raw material impurities, production equipment wear and tear, production process contamination, human error, and countless other sources,” Zhao continued.
“Today’s AI-powered vision systems have a remarkable ability to distinguish visually similar foreign materials on products, exceeding a level of accuracy beyond what most human inspectors can achieve.”
However, it’s important to note that these inspection systems are not designed to replace tools such as X-ray and metal detection systems.
“Each technology has its strengths: X-ray excels at detecting dense, embedded objects, while AI vision addresses surface-level soft materials that other technologies routinely miss. Together, they form a more complete inline food safety programme,” Zhao clarified.
AI vision in action
AI vision systems are now deployed around the world, spanning a wide range of food industries.
Potato sorting, sizing, and grading: Potatoes are typically sorted based on size, shape, and presence of defects. Every potato has its own customer, from the ‘fresh pack’ varieties that end in grocery stores and restaurants, to ‘process’ potatoes that are sliced and diced for chips and finished foods, and others, potato processors use AI to automatically sort potatoes into their ideal value streams.
Fruit and vegetable sorting and processing: AI systems help both fresh produce processors and individually quick-frozen (IQF) food manufacturers sort out unwanted foreign materials, especially materials that look like the product.
Meat, poultry and seafood processing: From detecting foreign materials from beef trim before it enters a grinder to analysing the breading coverage and colour of a chicken nugget, hygienically designed AI vision systems are instrumental in helping protein processors elevate their food safety, quality, and regulatory requirements.
Baked goods and snack foods: Some of the first AI-based vision systems were deployed in the baking and snack industry. From ensuring a bread dough is sufficiently proofed before baking, objectively measuring the size, shape, and bake colour of a bun so it meets specs to fit into packaging, to verifying correct filling levels in pies or pastries coming off a high-speed production line, there are expansive use cases for AI vision in this industry.
Cheese inspection: Shredded or shaped cheeses typically begin with a bulk block of cheese. It is quite common for packaging fragments to enter the process – especially clear plastic – but AI-powered inspection is uniquely trained to find these materials before they can reach later processing stages.
Finished food inspection: From counting the number of meatballs on a frozen dinner, confirming proper topping distribution on a frozen pizza, or spotting food caught within the plastic seal of a product container, AI vision is built to spot fine details often missed by human inspectors.
The drawbacks
While AI is generating a lot of buzz, there are areas one must consider and invest in if they want the best outcome.
Simply buying the technology is not enough, business also need to ensure they get and offer the right kind of training and get organizational buy-in.
Many people have different ideas on what AI inspection can do in terms of performance, but many do not consider the time it takes to successfully deploy the technology on their lines.
Yuegang Zhao, KPM Analytics, CCO, president of Lab Solutions and AI strategy expert
“Gathering product samples to train the AI model can take months, and even after the AI is at work on the line, frequent monitoring by an AI expert is necessary to safeguard performance,” said Zhao.
“Companies may want an out-of-the-box solution for their inspection system, but those who see long-term gains are the ones who remain patient during the training and deployment process.”
He continued: “There are also genuine technical boundaries to understand. AI vision systems can only inspect what the cameras can see – foreign materials embedded within a product, rather than visible on its surface, are beyond their reach.
“Environmental factors such as changes in lighting conditions, condensation, or significant variation in raw material suppliers can also affect model performance and may require retraining over time.”
Managing false rejection rates is another operational consideration.
“A system tuned too sensitively will reject good product alongside bad, creating its own production costs. Understanding these boundaries upfront is what separates a successful deployment from a frustrating one.”
As Zhao outlines, the evolution of AI is extremely fast pace; yesterday’s breakthrough innovation could be today’s dinosaur.
“This is why partnering with an AI vision system developer who has their finger on the pulse of AI innovations and new techniques is so important for food processors to maximise their investment and set realistic expectations,” he added.
What does the future hold?
For Zhao, he believes there are endless opportunities for advanced product measurements or foreign material models, but he also notes a growing interest in integrating AI inspection with other automated, physical systems.
“Integration with robotic systems has been used in fresh produce sorting, sizing, and grading for a few years now, but more food industries are exploring these applications,” he said.
“Some food companies have explored a multi-dimensional integration of AI-based vision with other inspection types like hyperspectral imaging or X-ray.
“Combining multiple inspection techniques can help companies achieve a more robust food quality and safety programme.”