About this Abstract |
| Meeting |
2026 TMS Annual Meeting & Exhibition
|
| Symposium
|
Advanced Real Time Imaging for Materials Science and Processing
|
| Presentation Title |
Utilizing Deep Learning-based Spatiotemporal Fusion to Develop High-Framerate and High-Resolution X-Ray Radiography for the High-Speed Imaging User Community |
| Author(s) |
Songyuan Tang, Tekin Bicer, Kamel Fezzaa, Samuel Clark |
| On-Site Speaker (Planned) |
Songyuan Tang |
| Abstract Scope |
In this talk, I will report the development of a new x-ray imaging capability to fuse two sequences of X-ray images complementary in framerate and spatial resolution, and reconstruct a target image sequence with high spatial resolution, high framerate, and high fidelity. This imaging capability features a Shimadzu ultra-high-speed camera, a Phantom high-speed camera, and a deep learning-based spatio-temporal fusion algorithm to fuse image sequences. When the algorithm was tested on an independent dataset collected from the additive manufacturing user experiment, with the two cameras operated at framerates of 1 MHz (pixel size: 3.82 µm) and 71.43 kHz (pixel size: 1.31 µm), the reconstructed image sequence can achieve a framerate of 1 MHz, pixel size of 1.31 µm, with a peak signal-to-noise ratio of no less than 30.50±0.09 dB. These results demonstrate the proposed technology can significantly increase the scientific throughput of a high-speed imaging experiment for the user community. |
| Proceedings Inclusion? |
Planned: |
| Keywords |
Machine Learning, |