A new method for classifying and segmenting material microstructure based on machine learning

The microstructural characteristics of materials determine their service performance. Therefore, the rapid identification of material microstructure and the accurate extraction of feature parameters are significant for the research and application of materials. However, most materials have diverse structure types and complex microstructures. With the gradual maturity of computer vision technology, it is increasingly being applied to studying material images. In this paper, a neural network-based material microstructure recognition and semantic segmentation model is designed to automatically identify and classify titanium alloy structures and then adaptively process images and extract features to overcome the challenges of efficient recognition and extraction of multiple structures of materials. The study completed the recognition of 2275 images of 15 types of titanium alloys through data set preparation, image preprocessing, model building, and parameter tuning, followed by image segmentation of morphologically processed images and labels based on U-net. Finally, connected domain computation successfully extracted the feature covariates in multiple structures of titanium alloys. This work demonstrates the application of data mining technology in metal microstructure image research and the implementation process. It completes the identification and characterization of the complex microstructure of the material.