Skip to content

Detection of surface defects with Deep Learning

Quality is the key to success. Until a few years ago, slight scratches in homogeneous surfaces were tolerated by customers. Nowadays, however, the customer expects a flawless, scratch-free surface. “Traditional” image processing quickly reaches its limits here because the defects occur in an infinite number of variants and therefore cannot be clearly described.

Deep learning or AI (artificial intelligence) in image processing is the key to success here. Eye Vision Technology has succeeded in integrating the complex functionality into an uncomplicated software command within the EyeVision image processing software. This means that you as the user do not have to deal with the complex artificial intelligence, but can adjust it to your application with just a few adjustments thanks to the homogeneity inspector.

A typical application of the homogeneity inspector is the detection of scratches in the stainless steel industry. If, for example, scratches and anomalies are detected in the raw material before a punching process, they can be omitted from the actual punching process so that only scratch-free punched parts are produced whose overall visual impression does not give rise to customer complaints. However, large-area abnormalities such as the structure of a surface can also be checked and qualified with the homogeneity inspector. The homogeneity inspector helps you to achieve error-free and therefore more efficient production.

Homogeneity (Surface) Inspector

For the automatic detection of defects on structured surfaces.

This can be used, for example, to

  • Surface defects
  • damage
  • impurities

dynamically and automatically.

Challenge and solution

The identification of defects on surfaces poses a special challenge. For example, defects on complex functional and aesthetic technical surfaces.

Deep learning provides a new solution to this difficulty. It works without predefining defects. And without teaching the test program.

Advantages of deep learning surface recognition

  • The algorithms can be integrated into any image processing software.
  • The tool can be used in any ROI that uses, for example, the size of the deviation as a quality criterion for the surface defect.
  • No pre-learning
  • No setting of parameters
  • the algorithm automatically adapts itself to any remaining surface
  • evaluation of the inspected surface in less than 50 ms on Core i3

Try out the EyeVision Software for free!