A UAV-based multi-defect real-time detection system for tunnel lining using attention mechanism-enhanced detection model

Traditional tunnel lining inspection methods often require costly specialized equipment and manual involvement, making them labor-intensive and inefficient. To address these challenges, this study proposes a UAV(unmanned aerial vehicle)-based real-time tunnel lining multi-defect detection system, leveraging the attention-enhanced TunnelScan model. Specifically designed for UAV-based inspections, TunnelScan incorporates a novel channel mixer-based attention mechanism module, GLUConv, which raises feature selectivity by amplifying relevant information while suppressing irrelevant noise. By integrating depthwise convolution, GLUConv reduces computational overhead and adapts effectively to the special spatial textures of tunnel linings. The model further utilizes a multi-scale feature pyramid network (ms-FPN) and a refined loss function, Focusing Slide Loss (FS Loss), to increase the detection accuracy and efficiency across varying defect types. The proposed model is validated using the UAV-Tunnel dataset, which comprises diverse images of tunnel lining defects and maintenance objects captured by UAVs. The results demonstrate that the model outperforms both baseline and state-of-the-art models in detecting tunnel lining defects. Furthermore, the UAV-based system enables real-time data collection and multi-defect detection, which not only generates maintenance reports but also minimizes manual intervention and ensures efficient navigation in complex tunnel environments. Extensive experiments carried out in real tunnel scenarios demonstrate the robustness and effectiveness of the system with TunnelScan, showcasing notable improvements in the efficiency and reliability of tunnel lining inspections.