Classification of soil types from GPR B Scans using deep learning techniques

Traditional methods for classification of soil types are time consuming, invasive and expensive. A non-invasive method like ground penetrating radar (GPR) provides a suitable way to classify soil types based on its electromagnetic properties. Deep learning algorithms have proven to be an effective tool for features extraction of GPR data. A deep convolutional neural network (CNN) model for automatic classification of soil types is proposed. A synthetic dataset is created using gprMax and used to train and validate the proposed CNN model. The proposed model shows good performance in classifying 7 different soil types from GPR B-Scan images. Upon testing the model on new and unseen data, its accuracy is found to be 97%.