Dynamic wave tunnel lining GPR images multi-disease detection method based on deep learning

Abstract

Ground-penetrating radar (GPR) is widely used in tunnel lining non-destructive testing. However, diverse diseases need more data, leading to low accuracy in multi-disease detection. This paper proposes a dynamic wave tunnel lining GPR images multi-disease detection method based on deep learning to improve the detection accuracy of multi-diseases. Firstly, the tunnel lining generative adversarial networks (TLGAN) model generates multi-dimensional GPR images. Secondly, we designed the Wave-backbone model according to the different disease shape attribute characteristics. Meanwhile, multi-dimensional space fusion combined with the global dynamic wave detection networks Wave-DetNets detects the captured multi-scale global waveform features. Finally, the disease depth was further estimated by detecting results. The experimental results show that the average model accuracy in this paper for identifying cavities, loost, and steel rebars is more than 91%, the omission rate reduces to 0.04, and the detection speed reaches 37FPS with a 320 × 320 GPR image sizing. The model has excellent stability and is better than the existing detection model.