Pattern recognition with machine learning on optical microscopy images of typical metallurgical microstructures.

Area of Science:

• Materials Science
• Computer Science
• Metallurgy

Background:

• Advanced materials characterization requires efficient analysis of microstructures.
• Manual examination of steel microstructures is time-consuming and prone to error.
• Automated methods are needed for high-throughput analysis in materials research.

Purpose of the Study:

• To demonstrate a novel and effective approach for pattern recognition in optical microscopic images of steels.
• To develop a reliable and automated method for segmenting steel microstructures using machine learning.
• To validate the accuracy of the automated segmentation against manual analysis.

Main Methods:

• Utilized a fast Random Forest statistical algorithm, a machine learning technique.
• Applied automated segmentation for identifying and analyzing typical steel microstructures.
• Compared segmentation results (percentage and location areas) with manual examination.

Main Results:

• The Random Forest algorithm provided reliable and automated segmentation of steel microstructures.
• Excellent agreement was observed between machine learning and manual examination results for microstructure areas.
• The technique proved effective for pattern recognition in optical microscopic steel images.

Conclusions:

• The developed microstructure pattern recognition and segmentation technique is highly effective.
• This automated approach can significantly reduce the time needed for analyzing large image datasets.
• The method supports efficient quality control and accelerates the discovery of new steels with desired properties.