Semantic Segmentation Framework for Automated Rock Type Identification in Geological Imagery

Lithological identification plays a critical role in mineral exploration and geological modeling, contributing significantly to ore system interpretation and regional prospectivity analysis. Conventional approaches often depend on expert interpretation and microscopic examination, which limit scalability and automation. This study introduces a semantic segmentation framework for automated recognition of multiple lithology types from geological imagery. A custom image dataset was constructed, comprising nine common lithological classes—including basalt, shale, diatomite, and others—acquired from core samples and outcrop photographs. To enhance model generalization, the dataset was expanded using data augmentation techniques such as rotation, mirroring, and color perturbation. The proposed method achieved high segmentation accuracy, with an mAP@0.5 of 89.6%, a recall of 92.1%, and a processing speed of 45 FPS under GPU conditions. The model demonstrated strong adaptability in handling blurred boundaries and fine-scale textures, which are common in geological images. This approach is well suited for lithological interpretation of drill cores, segmentation of remote sensing images, and automated identification of prospective metallogenic zones, offering a scalable solution for intelligent geological information extraction.