2025 Theses Doctoral

Towards Operational UAS-Based Landmine Detection: Vegetation, Validation, and Geophysical Application

Baur, Jasper

This dissertation addresses the decades old humanitarian crisis of landmine and unexploded ordnance contamination by applying modern advances in unmanned-aerial vehicles (UAV), geophysical sensors, and computer vision.

Chapter 1 focuses on quantifying the effect of vegetation on detecting surface ordnance. Potentially the most important consideration for UAV-based object detection in the natural environment is vegetation height and foliar cover, which can visually obscure the items a machine learning model is trained to detect. Hence, the accuracy of aerial detection of objects such as surface landmines and UXO is highly dependent on the height and density of vegetation in a given area. In this study, we develop a model derived from an area's digital surface model that estimates the detection accuracy (recall) of an object detection model as a function of occlusion due to vegetation coverage. This methodology has significant implications for determining the optimal location and time of year for UAV-based object detection tasks and quantifying the uncertainty of deep learning object detection models in the natural environment.

Chapter 2 develops the physical infrastructure required to advance landmine detection research by designing and creating a comprehensive and realistic seeded minefield with 143 diverse types of explosive ordnance. The field is designed to accelerate research in the field of humanitarian mine action by providing an accessible, diverse testing area located at OSU’s Center for Fire and Explosives, Forensic Investigation, Training and Research range in Pawnee, Oklahoma to address the lack of realistic accessible test sites for mine action researchers.

Chapter 3 introduces the anomaly (A), identifiable anomaly (I), and unique identifiable anomaly (U) AIU Index, which is an object detection disambiguation framework, that enables the comparison of different geophysical and imagery-based detection modalities while taking into account the detection fidelity and corresponding false positive rate. The AIU index prescribes essential context for false-positive-sensitive or resolution-poor object detection tasks such as landmine detection with applications in modality comparison, machine learning, and remote sensing data acquisition.

Finally, Chapter 4 builds on the efforts of Chapters 1, 2, and 3 by comparing UAV-based geophysical techniques for landmine detection with the AIU Index on the seeded minefield. This chapter encompassed the largest ever multi-modal remote sensing datasets on a seeded minefield and assessment on the state-of-art UAV-based explosive ordnance detection methods. Across 34 distinct datasets and 143 explosive ordnance items, we systematically assess state-of-the-art UAV-based remote sensing methods spanning visual, thermal, multispectral, hyperspectral, LiDAR, synthetic aperture radar, magnetometry, and ground-penetrating radar (GPR), alongside handheld and cart-based electromagnetic induction (EMI) metal detection, spectroscopy, and GPR. In conclusion, we found UAV-based RGB imagery was the most cost effective, scalable, and highest accuracy method for detecting surface targets, and the traditional handheld EMI-based metal detector was the most effective way to detect subsurface targets. This work aims to advance UAV-based landmine detection toward achieving tangible humanitarian impact in conflict and post-conflict regions.