An AI-Driven System Integrating Hyperspectral Data for Lithological Classification and RQD Assessment *N. Okada1, K. Takizawa1, S. Nakagawa2, U. Shinji3, Y. Ohtomo1, Y. Kawamura1 1Division of Sustainable Resources Engineering, Hokkaido University, Japan, (*Presenting author: okadan@eng.hokudai.ac.jp) 2Chi-ken Sogo Consultants Co., Ltd, Japan 3UGS, Japan ABSTRACT This study presents the development of a web-based application, IROMIE, designed to facilitate rock segmentation and Rock Quality Designation (RQD) measurement using hyperspectral imaging integrated with artificial intelligence (AI). Hyperspectral imaging provides detailed spectral signatures of minerals, enabling precise identification and classification of lithological units. However, traditional workflows often rely on manual interpretation, which is time-consuming and subject to operator variability. To address this limitation, we implemented an intuitive graphical user interface (GUI) that allows end-to-end analysis without requiring programming expertise. The application incorporates two original algorithms: (1) a rock segmentation algorithm that detects subtle mineralogical variations for accurate classification and (2) an RQD algorithm that employs automatic crack detection and spatial calibration for reliable structural assessment. These algorithms were implemented on an open-source Python platform, ensuring accessibility without license constraints. The system architecture combines TypeScript and Python for seamless front-end and back-end integration, with PostgreSQL for robust database management. Validation experiments were conducted using borehole core images and hyperspectral datasets. The segmentation algorithm successfully identified lithological boundaries and variations with high accuracy, including features that are difficult to detect through conventional visual inspection. For RQD measurement, the automated method achieved a mean deviation of 11.8 compared to manual measurements, corresponding to an approximate shift of three points on the Rock Mass Rating (RMR) scale. Importantly, the reproducibility and precision of the automated approach were superior to manual assessments, reducing subjectivity and improving consistency across different operators. By integrating mineralogical classification with structural evaluation, IROMIE demonstrates the feasibility of a comprehensive, non-contact rock mass characterization tool. Its adaptability allows users to build models using their own datasets, thereby overcoming the limitations of pre-trained systems and expanding applicability across diverse geological environments. The proposed framework not only enhances the accuracy and reliability of hyperspectral analysis but also democratizes access to advanced geotechnical evaluation, making it practical for researchers, engineers, and non-specialist users. Ultimately, this system contributes to more consistent, precise, and scientifically robust decision-making in civil
RkJQdWJsaXNoZXIy MTM0Mzk2