State-of-the-art image processing with artificial intelligence

Artificial intelligence has already long been used in our self-developed image processing in order to clearly distinguish genuine faults from harmless structures and thus to minimise the false rejection rate during the quality inspection. It is directly integrable into the compact HEUFT reflexx ^A.I. camera and ensures full detection accuracy.

Intelligent object detection instead of a simple brightness comparison during the image evaluation: HEUFT did not take this step just now but more than 20 years ago in order to increase the precision of the in-line inspection and to minimise the false rejection rate.

And our image processing has been using artificial intelligence (AI) for more than ten years in order to classify the detected objects, to distinguish real faults from harmless structures such as water drops, to implement a targeted teach-in and to reduce the proportion of falsely rejected good products to far below one percent.

Smart image processing now also in the new HEUFT reflexx ^A.I. camera

We have continuously optimised the HEUFT reflexx ^A.I. both on the software and on the hardware level. The smart image processing can even be integrated directly into the HEUFT reflexx ^A.I. camera which was developed especially for this purpose and which treats and evaluates its own images directly in real time. These do not have to be transferred to the respective checking or inspection device first. This raises learning-capable image analysis to a top level; the application possibilities of our modular systems reach completely new dimensions. Easily connectable to these, there is much more to be achieved in the high-speed inspection of empty and full containers.

And at a significantly higher resolution. For full detection reliability with a minimum false rejection rate, the hardware and software developed in-house combines classic image processing methods with modern AI techniques such as object recognition, classification and learning functions.

Targeted intervention instead of a black box

Does this leave you completely at the mercy of the AI, as if you were sitting in a self-driving car without a steering wheel and brake pedal? No! Because unlike other providers, the object evaluation does not come from an unchangeable black box with countless unknowns. Rather, humans can intervene in the machine learning in a targeted manner:

If, for example, an unknown object that has never appeared before is detected, the AI evaluates it as an error for safety reasons. However, the user always has the option of modifying the quality assessment based on his or her own experience and defining the identified structures as good and uncritical for product and packaging safety - or to have them temporarily tolerated if necessary. Without affecting the AI network or having to retrain it in a time- and energy-intensive manner, production can then continue safely even in such cases. In normal operation, everything continues to run fully automatically and highly securely.

More than just machine learning

If necessary, the assessment can be easily changed at any time: That is truly intelligent AI! It does not make the expertise of people with experience in filling and packing obsolete. Instead it flows into the smart image analysis when required. In this way, each individual object can be taught into the multidimensional feature space fully automatically and classified individually to distinguish real defects from harmless structures such as drops of water.

HEUFT reflexx ^A.I. therefore combines artificial intelligence with human and proven image analysis methods with modern ones in order to achieve perfect inspection results and to guarantee full product and packaging quality with unrestricted productivity.

Even now the image processing developed in-house – and therefore also the HEUFT reflexx ^A.I. camera – has considerably more to offer than just simple machine learning. And the AI development is going on continuously!