Peter Crozier1,Ramon Manzorro1,Yuchen Xu2,Joshua Vincent1,Roberto Rivera3,David Matteson2
Arizona State University1,Cornell University2,University of Puerto Rico at Mayagüez3
Peter Crozier1,Ramon Manzorro1,Yuchen Xu2,Joshua Vincent1,Roberto Rivera3,David Matteson2
Arizona State University1,Cornell University2,University of Puerto Rico at Mayagüez3
Information on the atomic-level dynamics of an evolving system is directly encoded in the coordinates and intensities of the atomic columns in electron microscopy images. However, times resolution in the millisecond regime, accessible in recently developed direct electron detectors, usually have low signal intensities (electrons per frame), thus leading to datasets severely degraded by noise. Consequently, tracking the dynamics of the system, i.e. position and occupancy of atomic columns, in time-resolved image series becomes challenging.<br/>Conventional algorithms to record the motion of atomic columns and retrieve their intensity often involve the identification of local maxima and fitting 2D Gaussian functions over the region of interest [1]. Such algorithms output precise results when the signal-to-noise ratio (SNR) is high but for time-resolved TEM series with low SNR, existing algorithms may provide inaccurate atomic coordinates and intensities. To improve the performance of current algorithms, we investigate an alternative, noise-robust blob detection technique established in the field of computer vision for identifying atomic columns in noisy TEM images [2].<br/>A simulated TEM dataset has been prepared and modified with the addition of high content of noise. Since the position and occupancy of each column is known (ground truth), the simulations allow a fair comparison between blob detection and other important algorithms in the electron microscopy field. The analysis demonstrates that, in the presence of noise, blob detection provides a more precise and more accurate fit on both the atomic column position and intensity. Given the promising results obtained from blob detection, the method has been applied to an experimental <i>in situ</i> TEM image-series of a CeO<sub>2</sub> nanoparticle undergoing dynamic re-arrangements [3].<br/>[1] Levin, B. D. A., et al. Ultramicroscopy, 2019. 213, 112978.<br/>[2] Manzorro, R., et al. Manuscript under revision.<br/>[3] We gratefully acknowledge support of NSF grant CBET-1604971, NRT-1922658, CCF-1934985, OAC-1940097, OAC-1940124 and OAC-1940276, and the facilities at ASU’s John M. Cowley Center for High Resolution Electron Microscopy.