Unsupervised Segmentation for Microstructure Identification of High Strength Steel with Superpixel Segmentation and Texture Feature Clustering

The mechanical properties of high strength steel are governed by its microstructure, which typically exhibits a multi-phase composition when the steel is highly strong and ductile. The distribution, proportion, and morphology of the constituent phases significantly influence the final mechanical properties of the material. In this regard, the development of an accurate and efficient method to identify and segment the microstructure of high strength steel is essential. To address this issue, a microscopic image database of high strength steel materials was established and an unsupervised image segmentation method that leverages superpixels and texture features clustering was proposed to identify the complex microstructure of high strength steel. By optimizing the number of superpixel divisions, our method achieved high accuracy in segmenting the three types of mixed structures in the dataset, with average pixel accuracy scores of 79.21% (M + B), 86.23% (F + P), and 84.32% (F + A). Furthermore, our method outperformed traditional unsupervised segmentation methods in segmenting microstructure images with intricate textures. The proposed method is thus deemed an effective tool for studying and applying high strength steel materials.