Investigating the microstructures of materials with microscopy is a key task in quality assurance, the development of new materials, and the optimization of manufacturing processes. However, conventional image analysis often demands significant time for analysis and a large volume of images, and the predictions produced are commonly constrained. Applying deep learning, models can be trained to analyze material microstructures quickly and with greater accuracy. The objective of this study is to provide a method for the automatic segmentation of microstructural images obtained from microscopes or scanning electron microscopes using Convolutional Neural Networks. For this purpose, two software scripts were developed in Python employing OpenCV and the fastai library. The first script is designed to generate reference images, while the second is utilized for training a model and predicting the microstructure in an image. The test of the microstructural analysis using the developed software tools demonstrates that robust prediction results are attainable by using high-quality reference images. This tool has been made available as an open-source on GitHub for public use in materials analysis and can be enhanced and further developed if required.
Read full abstract7-days of FREE Audio papers, translation & more with Prime
7-days of FREE Prime access