Speaker
Markus Reischl
Description
We show how to design an automated phase-analysis model based on a Convolutional Neural Networks (CNN). A framework for the efficient generation of simulated diffraction scans is developed, since real measured and labeled scans are hardly available. Using this synthetic database, a CNN is parameterized, trained and compared against the manual analysis. As a supportive approach, a denoising autoencoder is presented, to be used to eliminate background and other disturbing effects from the signal
Primary authors
Mr
Jan Schützke
Markus Reischl
