24.06.2019
2 minutes of reading
A new module is available on plug im!, IFPEN’s image processing platform, based on a new learning method: deep learning, a form of artificial intelligence.
It was designed as a result of collaborative work launched in 2017 with Dan Ciresan (Conderra Research, a former senior researcher at the Dalle Molle Institute for Artificial Intelligence). The collaboration led to the adaptation of a network of deep neurons and a methodology adapted to a limited annotated dataset, successfully employed for the detection of defects for four types of alumina catalyst supports. With this approach, each image pixel is considered as a clean sample, considerably increasing the volume of data available for learning.
This deep learning approach is currently being directly applied in various projects conducted by IFPEN’s R&I teams. Further updates will follow in a few months!