A method and system for physically constrained electron microscopy simulation generation and segmentation

Abstract

The application relates to a physical constraint electron microscope simulation generation and segmentation method and system, and relates to the technical field of computer vision and electron microscopic analysis. The method first adopts a simulation training data generation engine to construct a composite atomic structure model subject to physical constraints, synchronously generates a strictly aligned pixel-level segmentation mask, performs HRTEM imaging simulation and image enhancement based on the composite atomic structure model, establishes a simulation dataset in combination with the pixel-level segmentation mask, and establishes a real dataset based on real HRTEM imaging. Then, a simulation-to-real training process is performed, the segmentation model is pre-trained using the simulation dataset, and fine tuning is performed using the real dataset until the model training is completed. The model is applied to segmentation of input data. Without relying on a large number of high-quality pixel-level labels, the segmentation accuracy and training robustness of the segmentation model are effectively improved, the analysis efficiency of the segmentation model is significantly improved, and high-throughput and quasi-real-time applications are supported.

Images