Development of a Tetrahedral Magnetic Gradiometer for UXO and Landmine Detection

The global landmine and unexploded ordnance (UXO) crisis is a $200 billion dollar problem that causes 15,000-20,000 casualties annually. A terrain-adaptable, geophysics-based approach to detection is needed. Single-sensor unmanned aerial vehicle (UAV) methods often produce dangerous false flags, while current multi-sensor systems are land-based, expensive to manufacture, and cumbersome to transport to remote sites. Understanding how environmental conditions, host overburden lithology, and vegetation affect geophysical signatures of buried objects is crucial for making progress and reducing false flags. We are developing a multi-sensor package that can be integrated on a quad-copter unmanned aerial vehicle (UAV) that evaluates terrain and environmental conditions using machine learning, and weights on-board geophysical instrument(s) data to obtain the most accurate landmine detection performance. One of the techniques we are incorporating is magnetometery, which is often utilized in both ground and aerial landmine detection surveys. Slung-loaded total field magnetometers are commonly used in UAV-based surveys, but the lack of component information limits the ability to infer the orientation, distance, and magnetic moment of detected anomalies. Fluxgate magnetometers provide component information but suffer from calibration errors and noise when used in isolation. To overcome some of the practical limitations of a single fluxgate magnetometer, we have developed a magnetic gradiometer system consisting of four 3-component fluxgate magnetometers arranged in a tetrahedral configuration. By examining the gradients between each component of each fluxgate, we are able to streamline calibration efforts, and mitigate spatial variations that result from space weather and cultural noise sources. Here we present initial results from our magnetic gradiometer system for detecting inert landmines and proxy UXO in a controlled sand bed testing site and compare them to synthetic data. Finally, we discuss the practicality and utility of machine learning as a tool for anomaly detection and classification. We gratefully acknowledge the support of NSF Grant No. IIP2044611 and DoD NDSEG Fellowship.