Architecture Optimization and Interpretability in Neural Networks for HRTEM Segmentation

Abstract

Machine learning and deep learning are becoming key elements of Transmission Electron Microscopy (TEM) analysis. Several pipelines have been demonstrated both in the realm of high resolution STEM as well as TEM Most rely on a primary segmentation step with many using the U-Net architecture proposed by Ronneberger and coworkers However, a plethora of architectures have been proposed for segmentation of real images. There has been an effort in the biomedical community to understand the impact of network architecture on segmentation However, this area is still just beginning to be developed for segmentation of TEM for materials science

Ähnliche Arbeiten