Ground Penetrating Radar Underground Target Detection Based on GPR-YOLOv5

For the current Ground Penetrating Radar (GPR) image recognition problems such as low accuracy, this paper relies on the architecture of YOLOv5, ConvNeXt as the feature extraction network and proposes a new GPR image recognition network GPR-YOLOv5, and the common underground pipe corridors, underground cavities, galvanized water pipelines, PVC pipes and, etc were identified in the GPR images. The experimental results show that the recognition accuracy, average accuracy mean and F1 value of the GPR-YOLOv5 network are 93.21%, 92.46%, and 91.25%, respectively, when detecting and recognizing underground pipe corridors, underground voids, galvanized water transmission steel pipes and PVC pipes. Compared with YOLOv4 and YOLOv5 models, the proposed network has good recognition accuracy.