Tunnels are core structural components of modern transportation infrastructure, and their structural safety is directly related to the stability of traffic operations and the safety of public travel. Water leakage is a typical and highly prevalent defect in tunnels. If not detected and addressed in a timely manner, it can lead to corrosion of lining reinforcement, degradation of concrete mechanical properties, and in severe cases, even induce tunnel collapse disasters. However, the internal working conditions of tunnels are complex: uneven illumination distribution within the tunnel results in low contrast between leakage areas and the surrounding rock background; moreover, the morphology of water leakage is highly variable, making it difficult to extract effective defect features and substantially increasing the difficulty of defect target detection. To address these issues, this paper proposes an intelligent tunnel water leakage detection model based on an improved YOLOv10, termed LPM-YOLO. First, a Local-Global Attention Fusion Module (LGAFM) was designed to achieve adaptive weighted fusion of multi-scale features. Second, a Partial Token-Statistical Self-Attention (PTSSA) module was introduced, which overcomes the quadratic computational complexity of traditional self-attention and reduces it to linear complexity, thereby effectively improving the efficiency of long-range feature dependency modeling. Finally, drawing on the hybrid architecture concept of MambaVision, a C2f_MV module was introduced into the backbone network to enhance the model’s capability in capturing long-range spatial correlation features. Experimental results on a tunnel water leakage image dataset demonstrate that LPM-YOLO achieves 75.6% A₅₀ (average precision at IoU threshold of 0.50) and 41.7% A_50:95 (mean average precision over IoU thresholds from 0.50 to 0.95), which represent improvements of 6.7 and 2.5 percentage points, respectively, over the original YOLOv10-s. The proposed model can adapt to the complex imaging environment of tunnels, significantly improve the detection accuracy of water leakage defects, and provide reliable technical support for the intelligent defect inspection and digital safety operation and maintenance of tunnel lining structures..