Deep learning-based tunnel advance geological forecast for polygon targets via regional GPR data
Ting-wei Yang, Jing-he Li,*, Ye-hui Cao, Bin Xiong, and Liu-ye Wei
1. College of Earth Science, Guilin University of Technology, Guilin, Guangxi 541004, China
2. Guangxi Transportation Science and Technology Group Co., LTD, Nanning, 530007, China
Abstract：
With the increasing tunnel construction projects in China, geological hazards and construction accidents in tunnels occur frequently. The widely applied intelligent detection of ground-penetrating radar (GPR) for tunnel advance geological forecast (TAGF) rarely involves research and application exploration on irregular structures. The GPR dataset for TAGF in the Guangxi region meets the requirements for isomorphic source domain feature extraction. This paper proposes a methodology for creating polygon labeling of irregular structures and develops an image segmentation approach based on the Yolov5s deep learning framework, establishing the polygon-Yolov5s intelligent forecasting network for irregular geological hazards in GPR tunnel detection. Through case trials using both numerical and actual TAGF datasets, the feasibility and effectiveness of the proposed algorithm are validated by using the SSD algorithm and the traditional Yolov5s algorithm. The effective utilization of intelligent interpretation systems for irregular geological hazards would improve the efficiency and accuracy of geological forecasting and operational maintenance detection.
作者简介: Yang Ting-Wei graduated from Guilin University of Technology in 2008 with a B.S. in Exploration Technology an d Engineering, in 2011 with a M.S in Geodetection and Information Technology. He is currently pursuing an Ph.D in Geological Resources and Geological Engineering at the College of Earth Sciences, Guilin University of Technology. His current research focuses on AI theory and application in geophysics.
2025, Vol. 22 
