Predicting ptychography probe positions using single-shot phase retrieval neural network

TL;DR

This work addresses the challenge of large, accumulating probe-position errors in ptychography by introducing a two-stage strategy: a single-shot phase retrieval neural network (PtychoNN) predicts an image for each diffraction pattern, and robust image registration then recovers the complete scan path by solving a global linear system. The method demonstrates accurate probe-position prediction for errors up to about $10^2$ pixels, enabling high-quality reconstructions where conventional optimization fails, and provides a practical workflow for instruments lacking advanced position-control hardware. It further shows that predicted positions can initialize and accelerate traditional ptychographic refinements, with robustness to data scarcity and potential for integration with physics-informed learning. Collectively, these contributions offer a scalable, data-driven route to reliable ptychography in less-controlled experimental environments, broadening accessibility and throughput.

Abstract

Ptychography is a powerful imaging technique that is used in a variety of fields, including materials science, biology, and nanotechnology. However, the accuracy of the reconstructed ptychography image is highly dependent on the accuracy of the recorded probe positions which often contain errors. These errors are typically corrected jointly with phase retrieval through numerical optimization approaches. When the error accumulates along the scan path or when the error magnitude is large, these approaches may not converge with satisfactory result. We propose a fundamentally new approach for ptychography probe position prediction for data with large position errors, where a neural network is used to make single-shot phase retrieval on individual diffraction patterns, yielding the object image at each scan point. The pairwise offsets among these images are then found using a robust image registration method, and the results are combined to yield the complete scan path by constructing and solving a linear equation. We show that our method can achieve good position prediction accuracy for data with large and accumulating errors on the order of $10^2$ pixels, a magnitude that often makes optimization-based algorithms fail to converge. For ptychography instruments without sophisticated position control equipment such as interferometers, our method is of significant practical potential.

Figures (8)

• Figure 1: Workflow of our proposed probe position prediction method based on single-shot phase retrival and pairwise image registration.
• Figure 2: Comparison of the nominal positions, true positions, predicted positions, and positions after refinement initialized with predicted positions for all cases with accumulating errors in Table \ref{['tab:test_sets']}. On the top right of each subplot is the magnified sub-region around the graph's center.
• Figure 3: Ptychography reconstruction results of all cases with accumulating errors in Table \ref{['tab:test_sets']}. Column (a) -- (d) shows the reconstructed phase image using true, nominal, and predicted positions with or without further refinement using optimization-based algorithms. Case A1-1 -- A1-3 are most representative for cases with large accumulating errors as demonstrated by the sharp contrast between reconstructions using nominal positions (red dash-boxed images) and predicted positions (green dash-boxed images).
• Figure 4: Comparison of the nominal positions, true positions, predicted positions, and positions after refinement initialized with predicted positions for all cases with independent errors in Table \ref{['tab:test_sets']}. See the caption of Fig. \ref{['fig:random_etch_paths_accumulating']} for description of the figure's elements.
• Figure 5: Ptychography reconstruction results of all datasets with independent errors in Table \ref{['tab:test_sets']}. See the caption of Fig. \ref{['fig:random_etch_recons_accumulating']} for description of the figure's elements.