This study applies machine learning to identify microstructured phases such as ferrite, pearlite, and graphite in vermicular cast iron, addressing data processing challenges. These phases exhibit irregular shapes and heterogeneous distribution, complicating automated microstructure classification. Both supervised and unsupervised learning methods were evaluated for segmentation and classification. A key challenge was the limited training dataset, requiring augmentation techniques like rotation, mirroring, and brightness adjustments. Unsupervised algorithms, including KMeans and the Gaussian Mixture Model (GMM), struggled to differentiate phases with similar pixel intensities and textures, revealing their limitations in analyzing complex microstructures. For supervised learning, the Mask R-CNN network within the Detectron2 framework was used, with manually labeled data in the COCO format. Implementing this method involved integrating and processing large image datasets, extracting key features with local binary patterns (LBPs), and optimizing neural network parameters. The model achieved high accuracy in phase detection and classification, though precise segmentation of small regions remains challenging. This work highlights the importance of advanced AI algorithms in microstructural analysis and the need for further method development to improve accuracy with limited data. The presented approaches have broad applications in materials engineering, automating microstructure analysis and reducing subjective human errors. Research conducted as part of project POIR.04.01.04-00-027/18-00.