About this Abstract |
| Meeting |
2026 TMS Annual Meeting & Exhibition
|
| Symposium
|
Recent Advances in Electron Back-Scattered Diffraction and Related Techniques
|
| Presentation Title |
Deep Learning Application for Simultaneous Kikuchi and Spot Pattern Acquisition and Indexing SEM |
| Author(s) |
Yuxuan Zhang, Jakob Schiøtz, Håkon Wiik Ånes, Alice Bastos da Silva Fanta |
| On-Site Speaker (Planned) |
Alice Bastos da Silva Fanta |
| Abstract Scope |
Transmission Kikuchi diffraction (TKD) in the SEM can yield bright diffraction spots superimposed on faint Kikuchi bands under on-axis, low-exposure conditions. However, the intensity gap and the strong dynamical effects often force users to discard spot patterns and rely solely on Kikuchi bands to extract crystallographic information. We present a machine-learning and template-matching workflow that extracts and indexes both patterns from a single low-dose acquired image. A residual U-Net, trained on paired exposures datasets, reconstructs high-angle bands while preserving unsaturated spots; and a dual-path transformer separates the two signals enabling individual indexing. This pipeline drastically reduces electron dose, expands TKD’s applicability to beam-sensitive nanomaterials, and increases throughput for data acquisition. |
| Proceedings Inclusion? |
Planned: |
| Keywords |
Characterization, Machine Learning, Other |