Scanning Electron Microscopy combined with Energy Dispersive Spectroscopy (SEM-EDS) has long been a cornerstone in the microscopic characterization of cementitious materials due to its ability to deliver detailed structural and compositional insights. However, the requirement for sophisticated equipment and lengthy imaging procedures presents limitations, especially when exploring alternative, low-carbon compositions in modern cement systems. To address these challenges, this study leverages artificial intelligence, specifically Deep Learning, to develop a more accessible and efficient method for microstructure analysis using only SEM imagery. We propose a novel semantic segmentation framework based on the U-NET architecture, enhanced with well-established pre-trained backbones such as VGG16 and ResNet50 for feature extraction. This model aims to identify and delineate four key cementitious phases (clinker, portlandite, pores, and hydrates) directly from SEM images, without relying on chemical information typically provided by EDS. Multiple configurations were explored by varying the number of layers, filters, and convolutional kernel sizes to optimize phase detection performance. The primary challenge tackled in this work is the robust classification of microstructural phases using purely visual data, which demands precise pixel-wise segmentation in the absence of compositional information. Despite this, the architecture successfully demonstrated its capability to segment the target phases, showing strong preliminary results. Training was conducted on a dataset of Ordinary Portland Cement (OPC) samples, and model performance was evaluated using accuracy and Intersection over Union (IoU) metrics. While initial results are promising, ongoing refinement of both the model and training dataset is necessary to improve performance further. This approach illustrates the potential of Deep Learning in replacing traditional, resource-intensive analytical techniques. By enabling efficient and scalable microstructural analysis, it paves the way for AI-driven materials design, particularly in the pursuit of sustainable, high-performance cementitious systems, offering a transformative tool for both research and industry in the field of materials science and engineering.