Enhancing Landmine Detection Using Deep Learning: Leveraging AlexNet for Classification

Abstract

Landmine detection is the biggest challenge today, especially with the explosion of anti-tank and anti-personnel landmines, bombs, and unexploded ordnance. Many countries around the world are at risk from buried landmines. Various techniques automatically detect and recognize subsurface objects. However, they have limitations and must be more reliable when applied to all soils with different mediums to provide accurate results. Ground penetrating radar (GPR) uses electromagnetic signals to identify subsurface objects, and it detects landmines deeper with minimal metallic content than other sensors. Advances in deep learning techniques could revolutionize anti-personnel landmine detection and classification, demonstrating incredible results with high accuracy rates. The paper explores applying the deep learning approach (AlexNet) architecture using simulated GPR data to enhance the identification of landmines and similar underground objects. This study utilizes gprMax software to simulate the GPR data of buried objects. Subsequently, the data is implemented and analyzed within the MATLAB environment. Our findings demonstrate that AlexNet outperforms a standard convolutional neural network (CNN) model, achieving higher accuracy, precision, recall, and F1 scores in classifying metal pipes, metal tiffin boxes, and plastic tiffin boxes buried underground. It highlights the potential of deep learning for landmine detection using GPR data.