Defect images in auditing

Challenge

• Currently too much time is required to summarize the defect image (consisting of defect, type of defect and location on the vehicle) in the system
• Goals:
  • Faster and easier capture of the defect in the system by automated image recognition and classification of the defect
  • Generating a proposal list from which the matching attributes can be selected

Solution

Image recognition by Deep Neural Networks

• Creation and labeling of defect classes (manually)
• Training a deep neural network
• Generating proposals from the most likely matches
• Interactive (not fully automated)

Benefit

• Support of auditors in defect recognition; simplification of the recording process
• Faster recording of the defect image in the system without time-consuming manual search for the correct component and defect descriptions
• Uniform proposals for defect images ensure a consistent data basis