MSA-GAN: A novel method for inpainting EBSD image via cellular automation and deep learning

Electron backscatter diffraction (EBSD) is an important technique for analyzing the microstructure of materials. Unfortunately, EBSD images obtained by experimental means often suffer from different types of information loss, which affect the subsequent experimental analysis. Our study aims to study an EBSD image restoration method based on deep learning, which can solve the problem of various defects in EBSD imaging process. This paper proposes MSA-GAN: A multi-scale image restoration method based on generative adversarial networks (GAN) and feature pyramids (ASPP). To construct a dataset for training the model, a mask generation method based on cellular automata is proposed to simulate the information loss in EBSD images. The experimental results show that the MSA-GAN method can effectively restore large, medium and small types of defects in EBSD images, and achieve better results than other restoration methods, improving the quality and accuracy of the digital microstructure images. Our study provides a new technical means for the analysis of material microstructures, which has important theoretical and practical value.