A Drone-mounted Magnetometer System for Automatic Interference Removal and Landmine Detection

TL;DR

This work addresses the challenge of detecting ferromagnetic landmines from UAV-mounted magnetometers despite substantial interference from drone electronics. It introduces a compact frame-mounted two-magnetometer payload and a two-step WAIC-UP and RUDE pipeline, enabling wavelet-domain interference cancellation and unsupervised landmine anomaly detection, respectively. Monte Carlo simulations show WAIC-UP plus RUDE outperforms traditional low-pass filtering and thresholding across random mine placements and varying altitudes, achieving high recall and strong localization accuracy while reducing false positives. The approach offers a practical, low-cost solution for efficient demining with potential extensions to other UAV-based magnetic surveying tasks.

Abstract

Landmines have been extensively used in conflict zones as an indiscriminate weapon to control military movements, often remaining active long after hostilities have ended. Their presence poses a persistent danger to civilians, hindering post-war recovery efforts, causing injuries or death, and restricting access to essential land for agriculture and infrastructure. Unmanned aerial vehicles (UAV) equipped with magnetometers are commonly used to detect remnant hidden landmines but come with significant technical challenges due to magnetic field interference from UAV electronics such as motors. We propose the use of a frame-mounted UAV-borne two-magnetometer payload to perform a two-step automated interference removal and landmine detection analysis. The first step removes interference via the Wavelet-Adaptive Interference Cancellation for Underdetermined Platform (WAIC-UP) method designed for spaceflight magnetometers. The second method uses the Rapid Unsupervised Detection of Events (RUDE) algorithm to detect landmine signatures. This two-step WAIC-UP/RUDE approach with multiple magnetometers achieves high-fidelity ordinance detection at a low computational cost and simplifies the design of magnetic survey payloads. We validate the method through a Monte Carlo simulation of randomized landmine placements in a 10 x 10 m square grid and drone motor interference. Additionally, we assess the efficacy of the algorithm by varying the drone's altitude, examining its performance at different heights above the ground.

Figures (5)

• Figure 1: The left panel shows the UAV serpentine flight path over a 10 m x 10 m grid for landmine detection. The paths of the two magnetometers are shown in blue (Sensor 1) and green (Sensor 2). Red markers indicate landmine positions, which are randomly distributed and buried at depths between 0 to 15 cm. The right panel shows the layout of the UAV’s quadcopter motor setup. Each motor is displayed as a cylinder with a side length of 10 cm, with the magnetometers placed at the center of the UAV and 10 cm directly below the first magnetometer.
• Figure 2: Example spectrogram of the simulated magnetic field signals. The top panel shows the raw magnetometer signal composed of the UAV interference and landmine signals. The middle panel shows the signal cleaned by WAIC-UP. The bottom panel shows the true landmine magnetic field signal with no UAV interference along the same path.
• Figure 3: The top panel shows the raw absolute magnetic field signal from the bottom magnetometer in the UAV simulation. The middle panel shows the absolute magnetic field signal cleaned with the WAIC-UP algorithm. The bottom panel shows the absolute magnetic field signal cleaned with a simple low-pass filter. In the middle and bottom panel, the true landmine signal is plotted in the dashed green line. Anomalies identified by the RUDE algorithm are highlighted in red while nominal data is highlighted in blue.
• Figure 4: This box plot shows the error in meters between the estimated landmine positions and the actual center of the landmine for each interference removal and detection method. The red line shows the approximate radius of 15 cm of the M19 Landmine.
• Figure 5: This plot illustrates the performance of different combinations of landmine detection algorithms in terms of F1-Score (left axis) and correlation (right axis) as a function of UAV altitude. Four detection methods are compared: WAIC-UP combined with RUDE, Low-pass filtering combined with RUDE, WAIC-UP with Thresholding, and Low-pass filtering with Thresholding.