Intelligent Classification of Tunnel Surrounding Rock Based on Tunnel Face Image Recognition

Purpose—Accurate classification of tunnel surrounding rock is essential for safe excavation, timely support adjustment, and risk control during construction. This study proposes an intelligent classification method that links tunnel face image recognition with a modified BQ rock-mass quality framework. Design/methodology/approach—A YOLOv8 model was trained to detect and localize joints and fractures on tunnel face photographs. The detected fracture regions were processed through grayscale conversion, denoising, histogram equalization, adaptive thresholding, and morphological operations. Pixel statistics were used to calculate the fracture-area ratio and derive a crack ratio for evaluating tunnel face integrity. The image-derived integrity coefficient, rebound-test-based saturated uniaxial compressive strength, and correction coefficients were integrated to calculate the corrected BQ value. Findings—The method was validated using data from the Laimipo Tunnel in Yueyang, China. The YOLOv8 model identified tunnel face joints and fractures effectively, with a mean average precision (mAP@0.5) value of 0.898 for the fracture class. Field validation on ten tunnel faces showed that the predicted surrounding rock grades were consistent with the measured grades in nine cases, giving an accuracy of 90%. Originality/value—The study links image-based fracture recognition with the modified BQ method and provides a rapid, quantitative workflow for construction-stage surrounding rock classification and timely support adjustment.