For this client, Vision Partners developed a customer-specific product classification based on neural networks. An image is taken of the packaged pastry and the vision system validates that the correct box goes through the labeling machine.
For this client, Vision Partners developed product classification based on artificial intelligence:
- Creating an images dataset for training the network
- Checking the data set so that the training takes place only with the correct images
- Indicate properties of the products being trained:
- Excluding products that do not need to be recognized by the Neural Network
- Grouping of products in case certain items are too similar to each other
- Determine maximum number of items to be included in training
- Network training
- Assessment of the quality of network training using the "confusion matrix"
- Implementation of the trained network
Below is an example of a confusion matrix that validates how well the trained network recognizes different products.
Vision Partners developed the turn-key product classification and implemented it in the production line. The solution consists of:
- LED dome lighting so that glare-free recordings can be made
- Industrial machine vision camera placed in production line
- Control cabinet with machine vision processor in the server room
- Integration with the parent system. Both analog (for triggering) and TCP/IP (for communication of classification results belonging to the product ID).
- Development and release machine vision classification application
This Vision Partners product classification allows the customer to automatically classify the product much more efficiently.
"This prevents us from sending mislabeled products to our customers and allows us to focus our energy on getting more projects."
