Usage of U-Net and W-Net neural network architectures for steel samples metallographic analysis

Abstract

The article proposes an approach to the metallographic study of steel images based on the use of the W-Net neural network classifier. A software approach to data processing (micrographs of metal slices) has been developed, which includes image preprocessing, finding metal segments (grains), calculating their boundaries, areas and grain score, followed by constructing a histogram of the distribution of metal grain areas on micrographs. The efficiency of the proposed approach is analyzed by comparing the obtained histograms of distributions with the reference ones by calculating their statistical characteristics. The results show a good correlation between calculated and reference data.