The process recognition and cycle management stages of drill-and-blast railway tunnel construction are currently characterized by low levels of automation. To address this, an intelligent tunnel-construction analysis system that integrates sensing, recognition, and management is designed, developed, and validated to enhance the refinement and intelligence of construction management. Case studies are conducted in various tunnel-construction scenarios, with the Longquanshan Tunnel of the Chengdu-Chongqing Middle Line Railway serving as the primary case study. A complete automated system is established, which includes a two-stream deep-learning model based on an improved ResNet and a bidirectional long short-term memory network. The system enables the high-precision recognition of key construction processes by fusing spatial image features with construction temporal sequences. For easily confusable on-site procedures, a multilevel recognition strategy (coarse classification fine discrimination) is proposed, which improves the accuracy of fine-grained procedure classification. Furthermore, a construction cycle segmentation mechanism based on temporal logic is introduced to automatically partition complete construction cycles, provide accurate statistics on the duration and transition time of each procedure, and enable progress visualization and early warning of abnormal conditions. Experimental results show that the proposed model achieves an overall procedure-recognition accuracy of 94.7%, with a precision of 92.3%, a recall of 90.3%, and an
F1 score of 91.1%. Furthermore, the system performs cycle segmentation and duration analysis, thereby supporting automated construction-process management.
