To control the quality of the production process, Vision Partners has implemented an automatic classification system that supports the process operator to assess product characteristics such as shape, texture and color.
Deep-learning, locate anomalies
Deep-learning-based anomaly detection facilitates automated surface inspection for defect detection and segmentation, for example. The technology can flawlessly and independently locate anomalies, i.e. defects of any type, on subsequent images. Training the model requires only images of products without defects. Unlike other deep learning methods, no labeling is required. During run-time, anomaly detection segments areas of images that differ significantly from the training images.
AI (artificial intelligence) application
Vision Partners developed this client-specific artificial intelligence application using the Deep Learning Framework from MVTEc.
In addition to the turn-key hardware, the following software functionality was also provided:
- Image acquisition and display of multiple cameras on the operator console.
- Deviation detection product inspection results on operator console.
- Log images and inspection results in the Windows shared directory.
- Selection by operator of which product is produced.
- Training (by process engineer) of new recipes based on deep learning techniques.
Certified MVTEC partner
Vision Partners is an MVTEC Certified Integration Partner and used the Deep Learning Framework of MVTec. In addition to the proposed anomaly detection, the MVTec Deep Learning Framework also supports object detection, classification and semantic segmentation. To support these approaches, a Deep Learning tool is also available. The Deep Learning Tool integrates seamlessly into the MVTec product portfolio. After labeling, training and evaluation in the Deep Learning Tool, the trained network can be deployed in the respective runtime environment.
