Mayeryn AI Vision Classifier
Description
A proprietary AI system based on deep neural networks, designed for real-time defect detection and product classification. The system is scalable and parameterized, making it adaptable to practically any production line. It operates in visible light, with the possibility of expansion to IR/UV and 3D sensors.
The system can work as a classification system, selection system, quality control system, counting system, sorting system or grouping system, and measuring system.
Video presenting the operation of Mayeryn AI Vision Classifier
Video presenting the operation of Mayeryn AI Vision Classifier
For what materials?
The system is suitable for the classification and sorting of any product, including those difficult to photograph, such as glass, transparent plastic, shiny and reflective materials, as well as matte materials that absorb a very large amount of light, e.g., rubber, EPDM.
Key features
The system is suitable for the classification and sorting of any product, including those difficult to photograph, such as glass, transparent plastic, shiny and reflective materials, as well as matte materials that absorb a very large amount of light, e.g., rubber, EPDM.
Key features
- > Detection of damages, deformations, scratches, and geometric defects.
- > Classification of product types in industrial conditions.
- > Possibility to train models on new data.
- > Integration with PLC.
Specifications
Scalability
Mayeryn AI Vision Classifier is scalable thanks to its modular design. As a result, it can be used in simple applications that do not require high precision or processing speed, as well as in systems where hundreds of gigabytes must be processed every second, with precision at the level of hundredths of a millimeter and infallibility at the level of 99.95%.
Video presenting the operation of the line, which is based on Mayeryn AI Vision.
- > Number of processed images per second: 75 for 3 cameras (can be increased);
- > Element processing time: 5ms – 200ms (can be shortened);
- > Classification: damages, deformations, defects, condition, position (customizable);
- > Number of training base images: from 100 to 100,000 depending on the complexity of the problem;
- > Queuing and multithreading system;
Scalability
Mayeryn AI Vision Classifier is scalable thanks to its modular design. As a result, it can be used in simple applications that do not require high precision or processing speed, as well as in systems where hundreds of gigabytes must be processed every second, with precision at the level of hundredths of a millimeter and infallibility at the level of 99.95%.
Video presenting the operation of the line, which is based on Mayeryn AI Vision.
Equipment list
- > 2D FHD cameras;
- > Image processing server;
- > PLC + actuators (sorters, rejectors, signaling devices);
Additional improvements
- > Automatic error logging and work tracking;
- > Operator panel/WebUI (camera preview, parameter changes, statistics);
- > Active learning mode – the system itself asks the operator to add missing labels and adds them to the training database;
- > Anomaly detection (unsupervised), useful when defects are difficult to define;
Compatibility with various types of data input interfaces
The biggest advantage of the system is its flexibility. Although the AI models of the Mayeryn AI Vision Classifier system were originally designed to work with vision systems, it is also suitable for working with other data input interfaces such as:
- scanners (linear, 2D and 3D)
- X-ray (Roentgen)
- sonars (linear, 2D and 3D)
- lidars (linear, 2D and 3D)
- bi- and multispectral cameras
- thermal imaging cameras
- microscopes (including electron microscopes)
- microphones
- electrical parameter meters
