Ordnance detection and discrimination techniques are typically based on two-dimensional (2D) representations of geophysical data such as contoured grids. These grids are created by interpolating data collected along transect lines. Without the use of wheeled carts (which pose their own set of unique restrictions), the assumption that all data are collected at a common height above the ground is routinely violated. Differences in sensor height between adjacent lines cause systematic and pseudo-random distortions in the gridded data which are not based on geophysical changes. The resulting anomalies are easily misinterpreted, or require filters which reduce the reliability of inversion routines. Field efforts to collect data of the quality required for adequate inversion are painstakingly conducted to achieve a measure of uniformity. The time, effort, and training required by these efforts have seldom been translated into cost-effective project implementation.
The objective of this research was to determine some of the benefits accrued by collecting and processing three-dimensionally (3D) positioned data for unexploded ordnance (UXO) detection and discrimination versus traditional 2D approaches.
The technical approach of this project was strictly model-based in that no geophysical data were collected. Although the techniques are equally applicable to electromagnetic as well as magnetic data sets, only magnetic data were modeled at this stage. 2D and 3D collection scenarios were devised and tested over a variety of single and double target models. Simulated targets included 60mm and 20mm shells at relatively deep burial depths and close proximity. Moderate levels of noise were added to the positions and to the synthetic magnetic readings.
In this project, 3D processing, visualization, detection and discrimination techniques were developed as analogs to the 2D techniques most commonly in current use.
Comparisons were made between the 2D and 3D techniques to evaluate the relative merits of the 3D approach.
A new collection procedure was devised to maximize the utility of the 3D approach. This is principally a “reacquisition” or “cued investigation” approach in which an operator is directed to an anomalous location on a dig list and proceeds to collect data over the target area by waving the sensor over the center point. This is similar to the approach currently used by UXO Technicians with analog magnetometers, only in this case the instruments would be digital and capable of recording 3D positioned data. Eventually, the processing and inversion techniques developed here can be implemented on a realtime basis in order to minimize field time.
Processing techniques were extended to 3D using commercial software. A variety of “gridding” routines were tested for their ability to reliably interpolate potential field data and an optimal technique was chosen. Hanning filter coefficients were also calculated and programmed to smooth the data in 3D.
Visualization techniques are irrelevant to numerical inversion routines, but are required for operator input. For example, without a simple visualization technique the operator cannot tell whether to invert for one or more objects. A variety of standard visualization techniques were examined, but the most effective was the Gradient String. This is a unique concept devised specifically for this project. A Gradient String traces a line of maximum magnetic gradient through the 3D data space above a target. This reduces the solid cube of data into a single strand which points directly to the target or targets.
Discrimination techniques included a variety of 3D analogs to 2D techniques. For the purposes of this project, discrimination was differentiated from detection in that detection was limited to identifying the number of targets, whereas discrimination attempts to determine specific parameters such as depth, size and orientation. In particular, higher order gradients were used to estimate depths as was a simple Euler deconvolution. The Multi-Sensor Towed Array Detection System (MTADS)- Data Analysis System (DAS) dipole model code was used to estimate depth and size. The commercial ModelVisionPro software was used to estimate depth, size and orientation. Discrimination in this study did not include differentiating UXO from non-UXO.
The 3D acquisition and processing methodology was shown to be better than the standard 2D method for detection and location of closely spaced targets in all cases. The Gradient String approach detected all of the targets and located them in (x, y) more accurately than the comparable technique of 2D analytic signal peaks. Depth estimates using the analytic signal ratio and Euler deconvolution were five times more accurate using the 3D data.
The most accurate technique was the 3D ModelVision spherical inversion, which averaged location errors of 9cm and depth errors of 2cm.
Tests showed that detection and inversion of 2D noise-free data was comparable to the 3D noise-free data in most cases. This was not unexpected for potential field data. The introduction of noise (1cm of positional noise and 0.1nT of magnetometer noise) created significant problems for the 2D data. In general, the data were insufficient to invert or detect multiple objects even with tight line spacing in some cases. The 3D data also suffered from the noise, particularly at low heights where gradients were strongest, but detection and location capabilities were almost always two to five times better than the comparable 2D data.
The Gradient String technique proved to be an excellent method of detecting and visualizing multiple targets within the data space. The various inversion techniques generally improved the 2D results, but the 3D processing outperformed the comparable 2D processing in almost all cases. The 3D data provided enough additional information to improve detection and depth estimates by up to five times, but not enough to determine more complex parameters, such as dimensions and orientation.
Several lines of development are suggested by the results of this project. Implementation of feedback mechanisms and an extension of the 3D techniques to electromagnetic methods are recommended to extract the additional features such as orientation from the data. More advanced data segmentation techniques are also recommended in order to more effectively resolve model parameters. Discrepancies between the various software packages for ellipsoid model signatures were found and need to be resolved. Test-stand measurements to verify signature shapes and field trials to verify performance standards are required. Similarly, tests to improve inversion results for target shape and orientation and for discrimination between UXO and non-UXO are still required to fully implement this 3D technique.