• Classification of Radar facies
• Convolutional Neural Network
• GPR
• Transfer Learning

Ground penetrating radar (GPR) is one of the most widely used geophysical methods. GPR is a non-destructive method which gives a precise and accurate subsurface view. It is widely used in Civil Engineering, Archaeological, Mining and Geology. The interpretation of GPR images is still a complex process which demands domain knowledge and substantial manual labour. Through this thesis I have tried to address these issues by employing machine-learning techniques to automate the interpretation of GPR images, concentrating on the analysis of radar facies. My idea was to construct a Machine learning model for GPR image classification based on different radar facies. A dataset of radar facies images was prepared by collecting data from previous studies and research papers and was further enriched by image augmentation process to diversify the dataset. I have built my own neural network for classifying radar facies and later I have also used transfer learning approaches, pre-trained VGG16 and ResNet50 models were fine-tuned to classify and segment radar facies efficiently. All three models were evaluated based on performance metrics of Test accuracy. Compared to the traditional interpretation methods the present approach shows improved accuracy and efficiency. This research provides an intention how Machine learning can be potentially used in geophysical interpretation.