Like unstructured texts, images also carry a wealth of information. Whenever complex visual inspection procedures are needed for manufacturing processes that are time consuming or costly to program with conventional image processing, deep learning is used. Deep learning learns and identifies, from sample data, structures and patterns. This knowledge is applied to new, unknown manufacturing processes. Deep Learning uses a variety of sample images to learn what a good part looks like on a camera image. These self-imposed display variants are used by the system to detect and sort out bad parts. The machine learns in this way to distinguish bad from good. This main advantage over conventional image processing solutions is future-oriented. On the other hand, fluctuations and deviations in optically similar parts are difficult to recognize in current image processing solutions. The machine image analysis is also used for sorting or assembly work. Or to light-dark differences which comes to bear with surface structures.
Finally, it can be said that conventional image processing recognizes quality problems and product defects that humans themselves have programmed. On the other hand, with deep learning unexpected errors or product anomalies can be found and derived from new findings. This technology still has great growth potential in terms of quality and pattern recognition.
0 Comments