Image segmentation using deep neural networks to eliminate shift in laboratory sample images caused by thermal drift

Not scheduled
20m
80/1-001 - Globe of Science and Innovation - 1st Floor (CERN)

80/1-001 - Globe of Science and Innovation - 1st Floor

CERN

Esplanade des Particules 1, 1211 Meyrin, Switzerland
60
Show room on map
Poster Anomaly Detection and Diagnostics Poster session

Speaker

Mateusz Floras

Description

The National Synchrotron Radiation Centre SOLARIS, a third-generation light source, is the only synchrotron facility in Central-Eastern Europe, located in Poland. The SOLARIS Centre, equipped with seven fully operational beamlines, serves as a key research hub for a wide array of scientific disciplines. The Centre requires advanced software tools to support the analysis of experimental data and has developed a specialized application to assist scientists in interpreting their results. One significant challenge faced during research is correcting image shifts in laboratory samples due to thermal drift. An additional challenge is the varying contrast, which depends on the amount of energy absorbed by the tested material. While traditional phase correlation algorithms based on Fourier transform have not provided entirely satisfactory outcomes, SOLARIS has integrated deep neural networks to improve image stability and enhance accuracy. In particular, the U-Net model is utilized for segmentation, enabling the detection of elements in often blurry images. Model training is conducted using thousands of images generated from experiments utilizing X-ray radiation, ensuring high precision in data analysis.

Author

Presentation materials

There are no materials yet.