In the food sector, the customer wants a solution based on deep learning for cutting food products
After years of using rule-based algorithms, today’s solutions are more flexible, more reliable and easier to maintain.
Evolve rule-based platforms, which have become difficult to maintain, towards a generic solution based on deep learning.
• C++
• HDF5
• Python
• OpenCV
• Pytorch
LTI proposed the development of a framework simplifying development and maintenance by creating a generic machine concept that is reusable for all cutting machines. This allows the software to be reused for all machines. So only the parameters will need to be changed, and not the application code.
Image analysis tools using deep neural networks have been created in order to accomplish the detection, segmentation and classification steps required in real time by the systems, according to adjustable specifications.
This data-driven approach has its ecosystem of annotation and performance evaluation tools. It also allows modular assembly of processing units into a pipeline.
The amount of code to manage is reduced by up to 80% for certain features. The cut compliance rate for a system has been increased by 6% compared to previous methods. Software maintenance and configuration complexity are significantly reduced.
The development of new systems is accelerated by the ease of integration of new tools
