Application of Self-supervised Learning in Tunnel Lining Detection

Ground penetrating radar (GPR) is an important nondestructive technique in bridge and tunnel defect detection, and its signal intelligent analysis is a research hotspot. The radar data interpretation method using neural networks requires labeling a large amount of radar data. However, in actual work, there are problems such as sparse void sample data and morphological diversity. In this paper, the self-supervised learning method is adopted, and only a small amount of labeling data is needed to train the model parameters, which solves the problem of scarce radar data and large amount of labeling tasks, and improves the accuracy of identification of voids. By using different models to analyze the radar data, the results show that compared with the traditional supervised learning method, the self-supervised learning method has higher accuracy and stronger robustness, and has better industrial application value in the future.