Material Classification of Underground Objects from GPR Recordings Using Deep Learning Approach

Exploration and detection of underground objects without excavation can be achieved by utilizing ground penetrating radar. Since such an approach is nondestructive, electromagnetic radiation has been used in order to accomplish sub-surface surveying. The correct interpretation of acquired ground penetrating radar data can be demanding, time-consuming and very challenging especially when the observed environment is noisy. However, with the assistance of artificial intelligence algorithms, such data can be processed and analyzed at high speed and with high accuracy. The aim of this research is to develop a deep learning model for the material classification of underground objects from ground penetrating radar recordings. Within the recordings, the pipes are usually visually represented as hyperbola-shaped features of different characteristics. Annotated by experts and preprocessed, ground penetrating radar recordings are used as input to the deep convolutional neural networks. In order to estimate the performances of the models, stratified 5-fold cross-validation is utilized along with the area under the ROC curve and confusion matrix. The results showed that the best performing model architecture (EfficientNetB3) can be used for pipe material classification from ground penetrating radar recordings with satisfactory results.