Abstract:Vehicle mounted ground penetrating radar technology has been widely applied in the detection of subway tunnels, playing an important role in ensuring the safety and reliability of subway tunnels. In order to accurately detect the subway tunnel defects and improve the detection efficiency, the on-mounted radar detection system based on Yolov5 model is constructed. Signals and images were first denoised using zero time correction, deDC, background removal and image gain methods. Adopting the Yolov5 object detection model and introducing the SPP-Bottleneck module for improvement, a vehicle mounted ground penetrating radar detection system based on the Yolov5 model is constructed. The results show that the improved Yolov5 model has a higher F1 value compared to the original model under the same confidence level. In practical applications, the F1, accuracy, and recall average values of the vehicle mounted ground penetrating radar detection system based on the Yolov5 model are 0.884, 0.873, and 0.895, respectively. This model is effective for defect detection in tunnels. The Yolov5 object detection model has a detection time of 0.3 seconds, and its efficiency has been improved by 93.75%, 84.2%, and 50.0% compared to the other three detection models, which has more practical application value. This study has solved the problems existing in traditional vehicle mounted ground penetrating radar technology and is of great significance for the operation and maintenance of subways.