AI-supported machine vision
Thanks to image processing functions based on deep learning, excellent and consistent results can be achieved even under difficult conditions and unpredictable fluctuations in lighting, materials and products can be managed. Instead of writing fixed rules, a set of good reference images is uploaded. The deep learning toolchain integrated in Automation Studio simplifies setup with predefined training parameters. The elimination of complex manual configuration reduces errors, simplifies adaptation to new data and ensures flexible, efficient production. The B&R Smart Camera offers a range of powerful AI functions that are easy to set up without extensive knowledge of programming or image processing.
Optical character recognition (OCR)
Deep OCR can read distorted or low-contrast text that conventional rule-based systems would not recognize – preconfigured and ready to use.
Anomaly detection
Context anomaly detection learns to distinguish natural deviations from actual errors. An error database is not necessary for this, as the camera learns what is "good" from an example. It recognizes deviations from this independently.
Object recognition
Object recognition based on deep learning recognizes shapes, even if edges are blurred, the background is distracting or contours overlap.
Classification
Classification with deep learning can recognize product types or characteristics and can therefore sort products effortlessly, just as people distinguish apples from oranges.
Usage examples
OCR without training or parametrization
The integrated character recognition function has an exceptional capability to instantly recognize characters. There's no training or parametrization necessary to enable high-speed reading even with poor image quality or difficult characters, such as dot matrix fonts. A deep learning algorithm makes the function even more reliable and opens up exciting new ways to improve quality, boost productivity and prevent rejects – all while making manufacturing more flexible. All available Smart Camera features can be combined with deep OCR.
- World class performance of 26 TOPS
- High-speed reading even with poor image quality
- Recognize even difficult characters instantly
- ROI and confidence level are the only parameters
- Deep OCR for better results
Anomaly detection in a global context
Deep learning-based anomaly detection identifies even the smallest deviations in structure with high reliability. This detection technology has already been successfully tested for automatic quality control of wooden surfaces, textiles, welding joints and more.
Unlike other deep learning methods, it does not require specific labeling of different defect classes. In most cases, around 100 good images are sufficient to train the network. During inference, anomaly detection segments the regions of the images that differ significantly from the training images.
- Clearly highlighted surface defects in a heat map
- Relevance for quality can be finely adjusted with downstream rule-based algorithms
- Boosted machine performance without AI experts
- Higher productivity and quality with less scrap