Kombination von Bildauswerteverfahren und Neuronalen Netzen für die sichtsystemgestützte Oberflächenprüfung

A system for industrial inspection tasks by computer vision is under development within the project SIOB (Sichtgestützte Inspektion von Objekten). This project integrates knowledge-based methods (e.g. expert systems), image evaluation procedures, and artificial neural networks (NN) for classification purposes in order to obtain a flexible inspection system for different industrial application areas. As a representative example, the inspection of a conduction protective switchgear is considered. A flow chart of all components of the projected prototype is discussed and some relevant image processing methods for the extraction of features (edges, blobs) are described where by means of adaptive thresholds the effects of slow illumination variations could be avoided. Blobs are described by geometric moments which are the input for two different NN: The Backpropagation Network and the RCE (Restricted Coulomb Energy)-Network