Research on the Remote-Sensing Imaging Characteristics of Ground Penetrating Radar for Leakage from Buried Oil Pipelines in Loess Layers

Abstract

When oil contamination occurs in a loess layer, the high porosity of the loess causes rapid signal attenuation. Abnormal dielectric properties produce complex reflected signals, while vertical joints and fissures generate multiple reflections, making it difficult to resolve the details of oil contamination as clearly as in ordinary soils. To clarify the imaging characteristics of leakage from buried oil pipelines in loess, a physical model of buried pipelines under different leakage conditions was established. GprMax was used to simulate multiple leakage scenarios under practical operating conditions. A variety of analysis methods, including neural networks, reflected-wave feature extraction, and imaging-algorithm optimization, were adopted to mine imaging features related to pipeline leakage from the data. The results show that the YOLOv5 neural network has strong feature-extraction capability for forward-simulation images of oil-pipeline leakage and exhibits good generalization in leakage detection.