Quantitative analysis of the microstructure of Mg-RE alloys based on deep learning models

Microstructure is a key factor affecting the mechanical properties of materials, especially the morphology, distribution, and size characteristics of the second phase. Traditional conventional characterization methods do not yet have quantitative analysis tools. In recent years, the rise of deep learning technology, particularly breakthroughs in image segmentation and feature extraction, has provided new solutions for the automated analysis of alloy Microstructures. This study proposes a deep learning-based method for automatic identification and quantitative analysis of the second phase in Mg-RE alloys. By constructing a semantic segmentation model combining the U-Net structure with the CABM mechanism, we successfully achieved accurate identification of the second phase regions in the Microstructure of alloys. Not only did the Dice coefficient increase from 0.79 to 0.84, but the training time for each batch was reduced from 3000 to 480 ms. Using this model, we performed image segmentation and feature extraction on alloy samples under different heat treatment conditions, revealing the influence of solid solution treatment on second phase particles. In particular, the effect of solid solution treatment time on the size and distribution of the second phase was investigated, and the optimal solid solution time was determined through characteristic quantitative analysis. This study not only provides a new tool for the efficient analysis of alloy microstructure but also provides strong data support for optimizing alloy heat treatment processes, demonstrating significant academic value and application potential.