Imperfection recognition on lenses

Challenge

A company manufactures high-quality articles for the cosmetics industry. Specifically, it manufactures a plastic lens to optically enhance a cream container.

• Goals: 
  • Only objects of the highest quality are used: Without scratches, contaminations, inclusions etc.  
  • Quality inspections should be automated (despite imperfections being difficult for the human eye to detect)
• Challenge: 
  • Process differences of training vs. test data (lighting, background)

Solution

Image recognition by Deep Neural Networks:

• Installation of a camera and collection of image data
• Labeling images of defective products (manually)
• Training a deep neural network
• Implementation and integration of the quality assurance into the running production

Benefit

• Automation of manual, less “attractive” activities
• Increasing the quality of finished products