Publication

Abstract : The contamination of landmines and other unexploded ordnances has become a global safety problem. There is an urgent need to identify and demine the buried unexploded ordnance from the subsurface to save people and civilians from severe injury and death. Ground Penetrating Radar (GPR) is the best geophysical imaging technique for identifying and recognizing underground objects. It retrieves large GPR B-Scan datasets from the subsurface to be processed and classified using deep learning techniques. The proposed technique describes GPR B-Scan images of buried objects like landmines with casing materials made of metal and plastic, which can be classified using deep learning with attention mechanisms with MobileNetV2 as the base model. It uses deep learning architecture to successfully extract essential features and recognize them from GPR images of metal pipes, metal tiffin boxes, and plastic tiffin boxes. It includes creating novel datasets on gprMax software with the real-time scenario of a homogeneous medium, added roughness, water, and grass to support a heterogeneous soil medium with different relative permittivity, casing material, object size, and burial depth. The process initiates with the preprocessing of GPR data, applying data augmentation, splitting the datasets into training and testing, and building a deep learning architecture with attention mechanism; finally, training the network model and testing for GPR image classification. The multi-class classification evaluated and showcased improved performance using deep learning with attention mechanism than traditional methods. The developed model can potentially support new inventions in archaeology, infrastructure assessment, and explosive ordnance identification.