Abstract
Detecting buried unexploded ordnance (UXO) remains a critical global safety challenge. Traditional Transient Electromagnetic Methods (TEM) often struggle with "dense" targets—multiple small, shallow objects buried in close proximity—due to signal overlap and complex inversion math. This paper presents a novel imaging approach that bypasses traditional resistivity inversion in favor of direct secondary voltage imaging. By employing an improved watershed algorithm featuring adaptive saddle-point search (ASS) and background marking, the proposed method successfully discriminates between closely spaced targets, significantly reducing under-segmentation and over-segmentation errors.
Introduction
The detection of unexploded ordnance is a massive humanitarian effort, with millions of pieces of UXO scattered across over 100 countries. The Transient Electromagnetic Method (TEM) is a standard geophysical tool for this task, operating by inducing eddy currents in metallic objects and measuring the resulting secondary magnetic fields. However, when targets are shallow and densely packed, their electromagnetic signatures often merge into a single, indistinguishable "blob" in standard surveys. Traditional inversion techniques, which attempt to calculate underground resistivity, are often computationally expensive and unstable in these scenarios.
Methodology
To address the limitations of conventional TEM, the researchers developed a specialized hardware and software pipeline.
Small-Loop TEM System
The study utilizes a custom small-loop TEM system designed for high spatial resolution at shallow depths. Unlike large-scale survey coils, the small-loop configuration provides the precision necessary to capture the nuanced signal variations of crowded targets.
Direct Voltage Imaging
Instead of performing complex mathematical inversions to find "apparent resistivity," the method creates 2D distribution maps directly from the raw secondary voltage data. This approach is more stable and computationally efficient for real-time applications.
Improved Watershed Algorithm
The core of the imaging pipeline is a mathematical morphology technique known as the Watershed Transform, enhanced with two key features:
- Background Marking: Utilizing Otsu’s method (maximum between-class variance), the algorithm defines a clear background threshold. This prevents sensor noise and minor signal "burrs" from being misinterpreted as real targets, reducing false positives.
- Adaptive Saddle-point Search (ASS): When two targets are buried close together, their signals merge. The ASS algorithm automatically identifies the "saddle point"—the peak of the valley between two signal maxima—to establish a precise boundary, preventing the targets from being merged in the final image.
Experimental Validation
The approach was validated through both numerical simulations and field experiments.
- Simulation: A model containing five distinct targets (spheres, an ellipsoid, and a cuboid) buried at 0.5m was tested. While standard methods failed to separate the overlapping signals, the ASS-enhanced algorithm successfully identified all five individual objects.
- Field Test: Three UXO-like metal targets were buried at depths of 0.3m to 0.4m. Despite sparse sampling at 10cm intervals, the system correctly identified and separated the targets, even when their voltage contours were heavily overlapped.
Key Insights
The research highlighted several physical phenomena critical for UXO detection:
- Basin Effect: For very shallow targets, the signal directly above the object can be weaker than at its edges, creating a "basin" that must be accounted for to center the target accurately.
- Shell Thickness: There is an optimal shell thickness for UXO that produces the strongest electromagnetic response; signals drop off if the shell is either too thin (high resistance) or too thick.
- Target Attitude: The burial angle (attitude) of a UXO creates asymmetrical signal patterns, which the improved algorithm can leverage to distinguish between elongated ordnance and generic scrap metal.
Conclusion
The integration of small-loop TEM hardware with an improved watershed imaging algorithm provides a robust solution for detecting dense, shallow UXO. By focusing on direct voltage imaging and adaptive segmentation, the method offers a practical path forward for improving the safety and efficiency of demining operations. Future work will focus on refining depth and volume estimation and testing the system in more complex, mineralized soil environments.
Summary updated