<p>The accurate classification of tunnel surrounding rocks is essential for guiding construction schemes and optimizing support designs, thereby ensuring safety and efficiency. Traditional approaches rely heavily on manual judgment, which introduces significant subjectivity and requires parameter acquisition through offline tests, leading to cumbersome operational procedures. In addition, classification outcomes are typically segmented by continuous tunnel mileage, resulting in coarse results that fail to satisfy the demand for refined and dynamic evaluation of tunnel face rock. Within the context of drill-and-blast tunnel construction, this study introduced a classification method for tunnel face surrounding rock based on a hybrid CNN-LSTM model. Eight measurement-while-drilling parameters recorded by the drilling rig were employed as input features. Spatial correlations were extracted using CNN, while temporal variations were captured by LSTM, enabling the real-time and fine-grained prediction of surrounding rock grades across different tunnel face regions. Concurrently, a supporting classification system was developed, and inverse distance weighting interpolation was applied to visualize the classification results. The experimental evaluation demonstrated that the CNN-LSTM model achieved a classification accuracy of 98%, representing a 2%, 5%, and 3% improvement over the single CNN, LSTM, and MLP models, respectively. The developed system delivered real-time evaluation, straightforward operation, and rapid response, effectively satisfying on-site construction requirements. This approach realized the intellectualization and refinement of surrounding rock classification through data-driven modeling, successfully mitigating the subjectivity inherent in traditional methods. Moreover, the visualization results provide intuitive guidance for the dynamic adjustment of support schemes, markedly enhancing the safety and cost efficiency of tunnel construction.</p>