|
| | Abstract | | In recent years, the demand for the ground-penetrating radar (GPR) method has increased because of its simplicity, especially for shallow subsurface exploration in urban areas. Currently, rapid improvements in data acquisition technology have enabled more efficient collection of subsurface information via the GPR method. In contrast, the recognition of the data remains an exercise conducted by experts such as experienced engineers. To overcome this problem,"deep learning" technology has received attention in recent years. In deep learning, particularly supervised learning, labeled training data must be of adequate quantity and quality, which is the primary task of deep learning. In order to establish the datasets efficiently and sufficiently, we have generated learning data based on calculated data with an open-source simulator, gprMax, which simulates electromagnetic waves via the finite difference time-domain method, and tried a simple distinction by the dataset. |
|
|
|
| |
|