A robust method of denoising experimental micrographs using deep learning

Microstructure imaging is crucial in materials science, but experimental images often introduce noise that obscures critical structural details. This study presents a novel deep learning approach for robust microstructure image denoising, combining phase-field simulations, Fourier transform techniques, and an attention-based neural network. The innovative framework addresses dataset limitations by synthetically generating training data by combining computational phase-field microstructures with experimental optical micrographs. The neural network architecture features an attention mechanism that dynamically focuses on important microstructural features while systematically eliminating noise types like scratches and surface imperfections. Testing on a FeMnNi alloy system demonstrated the model's exceptional performance across multiple magnifications. By successfully removing diverse noise patterns while maintaining grain boundary integrity, the research provides a generalizable deep-learning framework for microstructure image enhancement with broad applicability in materials science.