Abstract: To address the problem that the interpretation of ground-penetrating radar (GPR) data for hidden defects in tunnel linings relies heavily on manual experience and is inefficient, an online intelligent interpretation system for the full-process analysis of GPR data was developed based on deep learning. The system adopts a browser/server (B/S) architecture and integrates three functional modules, namely data management, intelligent processing, and automatic defect recognition, enabling an end-to-end GPR data interpretation workflow. In the data processing stage, a generative adversarial network (GAN) based on Pix2PixHD was constructed to perform automatic denoising and enhancement of GPR data, thereby improving data quality. In the defect recognition stage, an improved You Only Look Once version 5 (YOLOv5) network was employed by incorporating an attention mechanism and a multi-scale feature detection structure, which enhanced the accuracy and robustness of defect identification. In addition, an interactive interface was implemented to support intuitive visualization of the interpretation results and to reduce the technical threshold for engineering users. The test results indicate that the proposed system achieves efficient processing and high-accuracy interpretation of GPR data, and it is applicable to practical tunnel engineering.