Unfavorable geological anomalies identification and location of tunnel geological prediction for ground penetrating radar based on YOLO-GTGP

During tunnel construction, tunnel geological prediction is a key technology for ensuring construction safety, optimizing construction plans, and reducing project risks. Among them, Ground Penetrating Radar (GPR) is a key means to identify geological anomalies in tunnel geological prediction, but it faces challenges such as low data interpretation efficiency, strong subjectivity, and insufficient adaptability of general algorithms. To address these issues, this paper proposes an innovative object detection algorithm, YOLO-GTGP (GPR Tunnel Geological Prediction). This detection model addresses the real-time requirements in tunnel scenes, the feature extraction and integration capabilities in complex scenes, and the challenges of overlapping object detection by designing GhostRepLite (GRL) and DualPath-CNNTransformer (DPCT) modules and introducing the Soft-NMS algorithm. Experiments have shown that the YOLO-GTGP algorithm achieved a Precision ( P ) of 93.4%, a Recall ( R ) of 93.2%, an mAP50 of 95.7%, and an mAP50:95 of 73.6% on self-built datasets, with significantly better performance than mainstream models such as Faster R-CNN, SSD, and YOLOv8, YOLOv9, YOLOv10, and YOLO11. In the GPR detection of two actual tunnel projects, the detection results are highly consistent with the professional report interpretation, and the predicted geological anomalies such as water-bearing areas are effectively verified in subsequent excavations, fully demonstrating the practical application value of the algorithm and providing reliable intelligent decision support for tunnel construction safety.