Machine Vision
Deep Learning
Deep learning and convolutional neural networks (CNN) learn defects that are hard to define with classical methods; instead of writing code for every new case, showing examples is enough.
The motivation behind deep learning in machine vision is to automate recognitions that are hard with classical image processing. Instead of describing every object by its features one by one, many examples whose class is known are given to the system; the system extracts their common and distinguishing aspects itself.
We use deep learning in projects such as fabric defect detection (oil marks, weft/warp misalignment, tears), finding defects in the natural texture of leather, and detecting chips on glass bottle rims. Example photos of each defect type are shown; training is done once and the running system decides in milliseconds.
In textile in particular we are developing a deep-learning-based machine to learn and find every type of fabric defect. In addition to classical methods, texture inspection and background estimation lift success rates considerably.
Highlights
- Artificial intelligence
- Classification
- Texture analysis
- In-house software, patented methods
Application areas
- Textile & fabric
- Leather
- Glass & bottles
- Metal casting
