By using electromagnetic signals, ground penetrating radar is a minimally invasive method to pinpoint underground utilities. Rapid urban expansion has increased the need for more underground infrastructure, which can often mean reconstructing existing underground utilities, such as water, sewer, gas, and fiber-optic lines. It is critical to know what already exists before starting excavation for new lines. Digging without knowing what is underground can result in unknown risks, for example hitting a gas line, which can have environmental impacts that affect the public but also can cause scheduling delays and added project costs. GPR data analysis is complex and requires an experienced person to interpret the data, which can be time consuming and expensive. 

Researchers Leila Carolina Martoni Amaral, Aditya Roshan, and Alireza Bayat wanted to find out if machine learning could be applied to detect and classify underground objects. In their research, they used a state-of-the-art machine learning algorithm to detect and classify underground objects using field-collected GPR images. The team trained the algorithm with a data set of GPR images coupled with a 2,000 MHz palm antenna. Their data set included five target objects, lengths of metal and plastic pipe, boulders, and air-filled and water-filled balloons (to represent air and water voids) in a lab setting. The potential for underground autodetection shown in this model is promising and could be improved with the addition of more GPR images to the training data set. Read more about their research in the Journal of Pipeline Systems Engineering and Practice at The abstract is below.


Ground penetrating radar (GPR) is widely used in subsurface utility mapping. It is a nondestructive tool that has gained popularity in supporting underground drilling projects such as horizontal directional drilling (HDD). Even with the benefits including equipment portability, low cost, and high versatility in locating underground objects, GPR has a drawback of the time spent and expertise needed in data interpretation. Recent researchers have shown success in utilizing machine learning (ML) algorithms in GPR images for the automatic detection of underground objects. However, due to the lack of availability of labeled GPR data sets, most of these algorithms used synthetic data. This study presents the application of the state-of-the-art You Only Look Once (YOLO) v5 algorithm to detect underground objects using GPR images. A GPR dataset was prepared by collecting GPR images in a laboratory setup. For this purpose, a commercially available 2GHz high-frequency GPR antenna was used, and a data set was collected with images of metal and PVC pipes, air and water voids, and boulders. The YOLOv5 algorithm was trained with a data set that successfully detected and classified underground objects to their respective classes.

Learn more about machine learning’s potential to locate and identify underground objects in the ASCE Library: