Generative diffusion modeling protocols for improving the Kikuchi pattern indexing in electron back-scatter diffraction

TL;DR

This work tackles the poor indexing reliability of EBSD Kikuchi patterns captured at high scan speeds by proposing conditional generative diffusion approaches to restore noisy patterns prior to indexing. It compares pixel-space diffusion with latent diffusion to balance restoration quality and computational efficiency, using conditioning on low-quality inputs to guide denoising. The methods significantly improve orientation indexing metrics (CI and PQ) for 1 ms and 0.5 ms exposures, bringing restored patterns closer to 100 ms ground truth, while leveraging DDIM and latent diffusion to achieve substantial speedups. The results show practical potential for real-time EBSD analysis at high throughput, with clear paths to further gains through domain-specific encoder fine-tuning and advanced diffusion architectures.

Abstract

Electron back-scatter diffraction (EBSD) has traditionally relied upon methods such as the Hough transform and dictionary Indexing to interpret diffraction patterns and extract crystallographic orientation. However, these methods encounter significant limitations, particularly when operating at high scanning speeds, where the exposure time per pattern is decreased beyond the operating sensitivity of CCD camera. Hence the signal to noise ratio decreases for the observed pattern which makes the pattern noisy, leading to reduced indexing accuracy. This research work aims to develop generative machine learning models for the post-processing or on-the-fly processing of Kikuchi patterns which are capable of restoring noisy EBSD patterns obtained at high scan speeds. These restored patterns can be used for the determination of crystal orientations to provide reliable indexing results. We compare the performance of such generative models in enhancing the quality of patterns captured at short exposure times (high scan speeds). An interesting observation is that the methodology is not data-hungry as typical machine learning methods.

Figures (24)

• Figure 1: Quality of Kikuchi patterns as a function of exposure time. A 100 ms exposure time is sufficient for the EBSD camera to capture all the fine details like lines and poles of the Kikuchi pattern clearly. Signal to noise ratio decreases for decreasing exposure time.
• Figure 2: CNN architecture mapping 3-dimensional output (Euler angles) to the input image of Kikuchi pattern. The notation used to mention the architecture is 'number of channels $@$ size of the features'.
• Figure 3: Comparison of true and predicted orientation maps. (a) Ground truth orientation map generated using MTEX with experimentally recorded Kikuchi patterns. (b) Predicted orientation map using CNN for 100 ms patterns. (c) Prediction on 25 ms pattern. (d) Prediction on 5 ms patterns. (e) Prediction on 2.5 ms patterns. (f) Prediction on 1 ms exposure time patterns.
• Figure 4: Predictive accuracy of CNN model (misorientation error vs exposure time). As the exposure time decreases, the misorientation error between true and predicted labels increases. It is evident that the CNN model performs well for the patterns up to 5 ms exposure time. But the predictive accuracy is significantly poor for the patterns obtained at short exposure times, e.g. 2.5 ms and 1 ms.
• Figure 5: The generative diffusion model-based pipeline for enhanced pattern indexing.