Semantic segmentation and quantitative analysis of AlB 2 microstructural evolution in Al–B alloys under varied heat treatments

Microstructural characterisation is essential for developing new materials and improving existing ones. This study presents an automated methodology for segmenting and quantifying AlB2 intermetallics in aluminium–boron alloy microstructures using deep learning. Al–B samples with different heat treatment conditions and durations of 5, 15, 30, 50 and 70 h were imaged, and AlB2 flakes were segmented via a DeepLabV3-based pixel-wise semantic segmentation model. The resulting masks enabled automatic extraction of key morphological features, including aspect ratio, area, phase density, circularity, eccentricity, and a homogeneity (D) index. The best-performing model achieved high segmentation accuracy, confirmed by F1-score, precision, recall, and IoU metrics. Human validation showed feature-extraction errors remained below 7%, demonstrating that AI-assisted microstructural analysis can provide rapid, objective, and reproducible results aligned with Industry 4.0.