Deep learning of crystalline defects from TEM images: a solution for the problem of ‘never enough training data’

Abstract Crystalline defects, such as line-like dislocations, play an important role for the performance and reliability of many metallic devices. Their interaction and evolution still poses a multitude of open questions to materials science and materials physics. In-situ transmission electron microscopy (TEM) experiments can provide important insights into how dislocations behave and move. The analysis of individual video frames from such experiments can provide useful insights but is limited by the capabilities of automated identification, digitization, and quantitative extraction of the dislocations as curved objects. The vast amount of data also makes manual annotation very time consuming, thereby limiting the use of deep learning (DL)-based, automated image analysis and segmentation of the dislocation microstructure. In this work, a parametric model for generating synthetic training data for segmentation of dislocations is developed. Even though domain scientists might dismiss synthetic images as artificial, our findings show that they can result in superior performance. Additionally, we propose an enhanced DL method optimized for segmenting overlapping or intersecting dislocation lines. Upon testing this framework on four distinct real datasets, we find that a model trained only on synthetic training data can also yield high-quality results on real images–even more so if the model is further fine-tuned on a few real images. Our approach demonstrates the potential of synthetic data in overcoming the limitations of manual annotation of TEM image data of dislocation microstructure, paving the way for more efficient and accurate analysis of dislocation microstructures. Last but not least, segmenting such thin, curvilinear structures is a task that is ubiquitous in many fields, which makes our method a potential candidate for other applications as well.