Automated, High Throughput Analysis of HRTEM Image Dataset

We describe a graph based algorithmic workflow to extract a large set of features from a large corpus of HRTEM dataset of a high performance organic semiconductor material system. The algorithm detects the crystal region in the image and further extracts various information from those regions like the crystal area, the orientation of the crystal and the exact d-spacing of the region. The framework is optimized to be near real-time, that can potentially allow incorporation into the instrument to make real time decisions on data sufficiency. We also demonstrate a Wasserstein distance-based method to determine data sufficiency, which acts as a stopping criterion for data collection.