Ms. Odenthal, why does your image analysis software 'VisionPro ViDi' rely on deep learning?

Cordula Odenthal: Deep learning is a subset of artificial intelligence and is used to predict patterns, detect anomalies or support decision-making. In industrial inspections, deep learning models can distinguish unacceptable errors and tolerate natural deviations in complex patterns. And they can be fully adapted to new examples without having to reprogram their core algorithms. 'VisionPro ViDi' learns what makes a 'good' image from a relatively small selection of sample images via deep neural networks and can thus, for example, analyze errors, locate and classify objects or read printed markings.

What is special about the image analysis tool?

Cordula Odenthal: It is the first deep learning-based image analysis software developed specifically for factory automation. It combines the scalability and reliability of conventional image processing with the innate ability of humans to recognize deviations. It is particularly suitable for complex inspections with large anomalies and unpredictable defects that are too numerous and complicated to program and maintain in a conventional vision system.

In addition, the deep learning-based 'ViDi' algorithms are optimized for industrial image analysis in the real world and require much smaller image sets and shorter training and validation times.

How is the software integrated?

Cordula Odenthal: The tool is used with 'VisionPro' or 'Cognex Designer', a graphical user interface that enables the creation of complex image processing applications via drag-and-drop, i.e. without scripting or programming. 'ViDi' has four different tools for localizing features, detecting anomalies and aesthetic errors, classifying objects or scenes and character recognition.

Do users need any special knowledge to work with the software?

Cordula Odenthal : A basic understanding of image processing and deep learning principles is essential. Programming skills are also required, depending on the software being used.

Cognex recently acquired Sualab, a Korean provider of deep learning-based image processing solutions. So Cognex is going deeper into the topic of AI?

Cordula Odenthal: For us, deep learning is a key technology that complements our core expertise in machine vision in a forward-looking way. The acquisition of Sualab, as well as our reinvestment of 14% of our annual turnover in research and development, show that we take our ambition to further develop the technology very seriously. There are still a number of inspection applications that are carried out by humans because automation has not been worthwhile or simply not possible.