The Difference Between Supervised & Unsupervised AI for Food Inspection

This article originally appeared in Food & Drink Manufacturing UK.

The excitement around AI-powered inspection is undeniable. However, it is essential for food processors to recognize that not all AI technologies are equal.  

How Do AI Food Inspection Technologies “Learn”?

While there are many nuances, there are two ways to train an AI food production inspection system. The first method is called “Supervised Training.” In Supervised AI, the system is shown several acceptable and unacceptable images of products to create an AI model. Anything from a bruise on a potato to paper on a beef trim line can become an AI model.  

With Supervised AI applications, a human AI training expert continually guides or corrects the AI in its decision-making. Over time, the AI begins to make quality assurance or foreign material inspection determinations independently, but a human element remains critical to the system.  

Because Supervised AI technologies make their decisions based on the data used to train models, they may struggle to detect anomalies or foreign materials not included in their training dataset. A skilled data scientist can train the model to detect the unexpected, but with a lower degree of accuracy.  

The other AI training method is “Unsupervised Training.” An Unsupervised AI is left to make its own decisions on product features, anomalies, or foreign materials without explicit examples or guidance. Unsupervised AI systems learn their criteria autonomously, which means they can adapt to process changes or foreign material variations over time. Nevertheless, their results may be unpredictable and risky.  

While one will rarely find a truly Unsupervised AI application in food product inspection, a processor may deploy its AI system without its operator receiving the proper training to correct the AI if its accuracy drifts. Undetected foreign objects could lead to costly product recalls.  

In some cases, an unsupervised model may be more capable of detecting unforeseen anomalies, however it comes with an increasing number of false positives, which create costly product waste.

Ask the Right Questions of Your AI Technology Provider

Be wary of technology suppliers that offer a plug-and-play, off-the-shelf AI platform. Training an AI inspection system is an iterative process that can take time depending on the complexity of the product feature or foreign material the operator wishes to detect. Every processing line – even within the same facility – is unique. Deploying an AI system and expecting it to instantly understand the products and various materials operators want to inspect or detect without a strong training program managed by an AI expert will likely produce undesired results.  

AI inspection technologies are key to helping food processors maximize capacities and address labor issues, but they are only as good as the support and training they receive.

‍