Adaptive partial scanning transmission electron microscopy with reinforcement learning

Jeffrey M Ede

doi:10.1088/2632-2153/abf5b6

Abstract

Compressed sensing can decrease scanning transmission electron microscopy electron dose and scan time with minimal information loss. Traditionally, sparse scans used in compressed sensing sample a static set of probing locations. However, dynamic scans that adapt to specimens are expected to be able to match or surpass the performance of static scans as static scans are a subset of possible dynamic scans. Thus, we present a prototype for a contiguous sparse scan system that piecewise adapts scan paths to specimens as they are scanned. Sampling directions for scan segments are chosen by a recurrent neural network (RNN) based on previously observed scan segments. The RNN is trained by reinforcement learning to cooperate with a feedforward convolutional neural network that completes the sparse scans. This paper presents our learning policy, experiments, and example partial scans, and discusses future research directions. Source code, pretrained models, and training data is openly accessible at https://github.com/Jeffrey-Ede/adaptive-scans.

Highlights

Most scan systems sample signals at sequences of discrete probing locations
We introduce a new approach to STEM compressed sensing where a scan system learns to adapt partial scans[16] to specimens by deep reinforcement learning[17] (RL)
The main limitation of our adaptive scan system is that generator errors are much higher when a generator is trained for a variety of scan paths than when it is trained for a single scan path

Summary

Introduction

Most scan systems sample signals at sequences of discrete probing locations. Examples include atomic force microscopy[1,2], computerized axial tomography[3,4], electron backscatter diffraction[5], scanning electron microscopy[6], scanning Raman spectroscopy[7], scanning transmission electron microscopy[8] (STEM) and X-ray diffraction spectroscopy[9]. In STEM, the high current density of electron probes produces radiation damage in many materials, limiting the range and types of investigations that can be performed[10,11]. Most STEM signals are oversampled[12] to ease visual inspection and decrease sub-Nyquist artefacts[13]. A variety of compressed sensing[14] algorithms have been developed to enable decreased STEM probing[15]. We introduce a new approach to STEM compressed sensing where a scan system learns to adapt partial scans[16] to specimens by deep reinforcement learning[17] (RL)

Methods

Findings

Discussion

Conclusion