Due to the large numbers (up to tens of thousands) of possible targets identified in nominal unexploded ordnance (UXO) surveys, efficient and reliable machine-aided target pickers should be used to identify targets for subsequent characterization. When selecting anomalies, the goal is to identify all anomalous features that may be caused by UXO, while minimizing operator time and eliminating operator bias. To facilitate advanced physics-based modeling, however, the target pickers should also be able to select data appropriate to the target, i.e., to outline or estimate the anomaly’s spatial extent. The current approach to target selection is either manual identification or amplitude thresholding. The former is time-intensive, not clearly defined, and prone to operator bias. The latter is sensitive to noise and is prone to over- or under-picking unless judicious oversight is exercised. Neither approach provides measures for estimating the footprint of the anomaly. The impact to the Department of Defense (DoD) is obvious. Systematic, fast, and robust target pickers can save money and produce a defensible target list compared to the current methods.
This project evaluated four automatic target pickers as well as the manual method and transitioned them to the user community via Oasis montaj™ by building custom Geosoft Executables (GX). Oasis montaj™ is a geophysical data processing and visualization package developed and marketed by Geosoft Incorporated. The four automatic target pickers were: (1) a wavelet-based detection algorithm, (2) clustering positive and negative peaks, (3) a dipole-based matched filter (MF), and (4) analytic signal (AS).
The demonstration was broken up into two phases. The first phase used a 60-dipole synthetic dataset to explore the parameter space and optimize the algorithms for the four automatic target pickers. The result of Phase 1 was a set of starting parameters that was used in Phase 2. The second phase applied the target pickers to seven magnetic datasets using the parameters output from Phase 1 as a starting point. The seven datasets possessed different signal and noise characteristics and anomaly densities. Three datasets provided from the U.S. Army Corps of Engineers (USACE)— helicopter-towed wide area assessment (WAA) dataset, a vehicle-towed transect WAA dataset, and vehicle-towed Multisensor Towed Array Detection System (MTADS) datasets from the Aberdeen Proving Ground (APG) standardized test site and Target S1 at Isleta Pueblo in New Mexico—were used for the evaluation. Because each dataset has its own unique data characteristics, the starting parameters were adjusted iteratively to achieve the best performance. The knowledge gained from Phase 1 was used to guide these adjustments.
The primary performance objectives were to detect greater than 90% of the ground truthed targets for production surveys (greater than 80% for WAA) in less than a quarter the time needed for the manual method. The detection objective was met by some of the automatic pickers on some of the datasets. The time objective was met by all the automatic methods for all the datasets.
The detection rates for the automatic methods were heavily dependent on the data quality, noise characteristics, and target density. In general, the amount of geology and background noise affected the performance of the automatic methods the most. All the methods require a threshold to be set that is based on the noise characteristics of the data. Because of this, automatic methods will pick anomalies with signal amplitudes above the threshold that are caused by noisy data or geology and not pick the valid anomalies located in the quieter areas that are below the threshold. Datasets that contained a consistent background noise level across the area, as seen in the Isleta dataset, performed well, with all methods exceeding the 90% detection rate. Conversely, the WAA helicopter dataset contained a variety of noise levels due to geology, which caused 50% detection rates for all automatic methods. The manual method has an advantage in these areas because the analyst can set a lower picking threshold and dynamically filter out the picks that are obviously due to geology. This ability may also be a drawback because it may introduce operator error or operator bias. The other main drawback to the manual method is the time required to pick the anomalies.
Of the automatic methods, the AS and wavelet method gave the best results overall, but each method had its own strengths and weaknesses, and the best method was very data-dependent. A general observation for all the automatic pickers is that they should not be run blindly. The analyst should carefully choose their parameters and analyze the results. The process of iteratively changing parameters and visual review of the results was essential in selecting the best parameters.
Implementation of the automatic picking methods used in this demonstration should considerably reduce the time and thus cost required to pick anomalies when compared to the manual method. The amount of cost savings will depend on the data. In areas with isolated anomalies and low background noise or geology, the cost savings will be maximized because the automatic pickers are able to detect over 90% of the anomalies in a fraction of the time compared to the manual method. As the geologic noise increases or data quality decreases, the cost savings will diminish because the missed anomalies will need to be filled in using the time-consuming manual method, but cost savings still should be significant.
The stakeholders and end users of this data processing and analysis technology include private contractors who conduct geophysical investigations in support of UXO cleanup programs and governmental employees who provide technical oversight. This demonstration showed the data products associated with this analysis approach and the inherent transparency of the target picking process. This basic information will help to improve the results of future geophysical investigations conducted by others. The market for this type of guidance document includes all practicing geophysical service firms currently working in the UXO industry.