In homogeneous products such as jam, they can still be easily detected on X-ray images using conventional image processing: Tiny foreign bodies such as glass splinters, stones or pieces of wire. The situation is different when they are hidden in structured products – for example in porridge or loosely packed almonds. In order to identify them reliably even in such a difficult environment and to distinguish them clearly from product structures and inhomogeneities an AI for X-ray image analysis which is capable of deep learning is already required. HEUFT reflexx A.I. will soon have it to offer as standard!
It is not only applied in order to free the images generated with unique X-ray flashes from image noise and artefacts. Using multi-layered neural networks, the intelligent image processing even finds minimal shape deviations, so that the detection of dangerous foreign objects is also possible where it was impossible until recently: in products that appear very inhomogeneous on the X-ray image, with structures of varying absorption and cavities between the individual products.
AI-assisted X-ray image processing for increased detection reliability
This increase in performance could be experienced at interpack 2023 in Düsseldorf not only with the new HEUFT eXaminer II XT prototype which detected glass splinters, stones or even pieces of wire of the same size during the pipeline inspection of unpacked product mass irrespective of whether they were currently in jam, porridge or almonds with the help of the latest HEUFT reflexx A.I. expansion stage.
A separate Enlightenment station at the HEUFT stand which illustrated the new possibilities of the intelligent image processing from our own development and production using the example of criss-crossing pasta products provided even more amazement for the visitors to the exhibition. The AI not only accurately found and marked the one straight pasta among the many ring-shaped pasta. It even reliably identified ring-shaped wire in real time that was virtually indistinguishable in shape and size from the good products.
Thanks to novel deep learning functions the HEUFT reflexx A.I. will therefore achieve considerably more reliability in the detection of unwanted foreign objects in the near future. The latest version of intelligent image processing is superior to both classical analysis procedures such as grey value and contrast detection and conventional machine learning particularly on X-ray images of an unordered quantity of structured individual products. It increases the already impressive precision of the pulsed X-ray inspection with foreign body detectors from the HEUFT eXaminer II series correspondingly strongly. And it reduces the false rejection rate.
