Deep learning with the new EyeVision 4.0 software

Deep learning is always used in image processing for optical quality assurance when the object to be inspected shows different variations that are difficult to grasp. An example of this is recognizing license plates and reading the characters. Certainly there are license plates on all cars, but these are not always attached to the same place or have different character sets. Another example is crack or scratch detection on components of any shape. With these two applications, the human recognizes errors immediately, but the classic image processing does not. However, deep learning offers the solution to this problem. Patterns are taught into a neural network and then recognized, even if they appear in front of the camera in different positions or even partially destroyed. This guarantees detection by a monitoring system even under non-optimal conditions. The classic evaluation is retained. Because just as a person can recognize a pattern well, he fails when it comes to the question of the exact dimensions of the pattern or the components. Here, the neural network uses the classic image processing tool for measurement.

The combination of deep learning and classic image processing is therefore also superior to a person, because a neural network is fatigue-free and always as precise from the beginning as the brain, which significantly throttles its recognition performance after just 15 minutes.


The extended deep learning components of Eye Vision 4.0 enable easy solving of complex recognition tasks. The innovative software learns from experience and understands the world in terms of a hierarchy of concepts. The hierarchy of the concepts allows the computer supporting EyeVision to learn complicated concepts by simply putting them together. Eye Vision Technology enables you to integrate deep learning through two different options: The deep learning library of the new EyeVision 4 software can be filled with pre-trained networks or with fully trained networks. The pre-learned networks require less qualified images than the fully learned networks. This makes the learning process faster with previously learned networks. Pre-learned networks are ready for use after a short time.

The fully learned networks are individually adapted to you and the tasks to be solved. The images required for this are trained by us exclusively for you in artificial neural networks. The EyeVision 4 also contains a number of new deep learning and machine learning functions, and numerous pre-learned networks are also integrated.

Spread the love

Leave a Reply

Your email address will not be published. Required fields are marked *