English：

Current methods for evaluating tunnel surrounding rock integrity often suffer from strongsubjectivity and high information acquisition costs, leading to potential misclassification of rock qualityand mismatched support parameters, resulting in unnecessary economic losses. To achieve rapidclassification of surrounding rock grades during tunnel construction and improve evaluation accuracy, thisstudy employs the YOLOv8 convolutional neural network model and image processing techniques to assessthe integrity of tunnel faces. The results demonstrate that the YOLOv8 deep learning algorithm caneffectively identify and localize joint fractures on tunnel faces, exhibiting high recognition accuracy, lowrecall rates, and fast processing speeds. The model’ s predictions align well with actual tunnel faceconditions. By integrating the YOLOv8 convolutional neural network model with the modified BQ valuemethod, this study evaluates the surrounding rock grade of the Yuanguping Tunnel, achieving aprediction accuracy of 90%, which closely matches the real conditions. The proposed method proveseffective in assessing tunnel face surrounding rock integrity and demonstrates promising practicalapplicability.