Description: In this research domain image signals from different camera sensors in visible and infrared spectral range are used for quality control in production processes. Current research involves the training of recognition systems with an unbalanced training data set. The focus is on algorithms such as support vector machines/data description and probability density-based methods. The background to this is that it is difficult up to impossible to produce defective copies of a product for training purposes. This is especially true for rare sporadic errors for which it is even doubtful whether an artificial generated error is adapted to reality. For this reason, the class of errors is often underrepresented and sometimes not even available. The basis for this research are formed by datasets of laser brazed joints, which were obtained in the project "Development and qualification of an on-line quality monitoring system for the industrial laser brazing". Further research issues include the integration of a reject decision level in detection systems and the implementation of fast algorithms for the online capability.
This research focus is placed in the areas of pattern recognition and image processing.