AI (Artificial Intelligence) makes quality inspections via image processing not only better, but also easier, faster and more efficient to implement. Data Spree from Berlin shows us how.

In the food industry in particular, efficiency along the entire value chain is an essential competitive factor. Reliable automation of manufacturing and quality assurance processes is crucial for a modern and efficient factory.


Classical image processing has to be programmed from scratch in a very complex way. Algorithms are developed manually by experts, which often requires a lot of know-how and time. At the same time, complex tasks, such as different or difficult defect images, cannot be mapped at all or only with great difficulty using these classic solutions. All this leads to high costs and to the fact that quality requirements often cannot be completely fulfilled.

This is why the food industry must use Vision AI in the future

annotation or labeling, takes place. The Data Spree actively supports here with annotation tools    and    services. Finally, the AI iteratively trains the recognition and correct assignment of the “Ok” and “Not Ok” examples. The AI works on the basis of an interconnection of nerve cells similar to the human brain. Here, the AI independently learns to distinguish good from bad products based on image data. As with the human brain, the accuracy of the AI is continuously improving. With Deep Learning DS, you can quickly and easily perform this “learning process” yourself. Data Spree also offers the complete process up to productive integration into the plant as a service.

This method can be used to quickly detect a wide variety of complex defect patterns, such as various surface defects, cracks, fractures, color defects and much more – and all without a single line of programming code. Quality assurance solutions can thus be implemented very efficiently and robustly. Even in just a few hours, operational prototypes can be created in some cases. Data Spree’s fast AI models additionally ensure good real-time capability in high-frequency production operations. Another advantage is the flexibility of the learning system. If products, product features or defects change at some point due to production changes, the AI can be easily “fed” with   new images and retrained. In this way, it is possible to react quickly and effectively to changes in production without having to start from scratch or buy a new solution.


In the quality assurance process of cookie products, the AI can reliably distinguish “Ok” from “Not Ok” objects. The output shows the result and a heatmap. The red area in the heatmap is the AI’s basis for deciding whether an object is classified as “Ok” or “Not Ok”

Here, the AI can easily detect a wide variety of defect patterns and deviations from the “Ok” state. The AI detects and localizes obvious defects such as holes, large cracks or fractures. But also more difficult defect patterns, such as small fractures and spalling or surface defects. The trained AI can independently detect and localize every possible defect variant or variable deviation, even if the defect did not explicitly occur in the training data set.

With AI from Data Spree, all possible error cases are thus detected here. In the past, this quality monitoring would have required complex algorithms to be programmed by hand to detect a wide variety of deviations and defect types. With Vision AI from Data Spree these times are over. Especially for tasks with high error variability, learning AI systems are excellently suited. This means that especially in the quality inspection of food and baked goods, it is always worth taking a look at the topic of AI.