A U-Net-based inversion model for underground cavity detection

Identification of the size and depth of underground cavities is essential for assessing underground hazards. This study proposes a deep learning-based inversion model that combines CycleGAN and U-Net models to identify underground cavities. First, CycleGAN is used to convert measured data into pseudo-simulated data, enhancing the visibility of cavity-related features in the images. U-Net is then employed for hyperbolic image segmentation, extracting detailed information about the upper and lower boundaries of the cavities. Based on the segmentation results, geometric parameters such as the central depth and radius are calculated, enabling accurate inversion of underground cavity characteristics. The model is validated using radar data obtained from laboratory experiments. The results show that the proposed method can accurately identify the depth and size of cavities. In addition, CycleGAN effectively transforms radar images into feature-enhanced representations, facilitating the detection of underground cavities.