Accurate identification of rock mass fractures is essential for evaluating tunnel stability and ensuring construction safety. However, existing deep learning-based approaches frequently exhibit limited performance in challenging tunnel environments and often lack integration with key engineering classification standards, thereby offering insufficient direct support for engineering decision-making. To address these issues, this study introduces an intelligent framework that integrates an improved YOLOv8-seg model with a standardized rock mass classification system. The improved model incorporates an Efficient Channel Attention (ECA) mechanism, which substantially enhances the detection of fracture features under complex conditions. Furthermore, the framework automatically extracts key fracture parameters, including fracture width and filling state. These parameters are directly correlated with the rock mass classification criteria specified in the Chinese National Standard GB 50487-2008. Experimental results demonstrate that the proposed method achieves a notable improvement in detection accuracy, evidenced by a 1.80% increase in mAP@0.5. At the same time, it retains real-time processing capabilities. The automated classification outcomes exhibit strong consistency with manual expert assessments, providing a reliable and efficient tool for supporting engineering decisions in tunnel construction.
Wang et al. (Sat,) studied this question.