For mobile, landscape view is recommended.
Results from recent ESTCP live site demonstrations have shown significant improvements, relative to only a few years ago, in the ability to accurately distinguish between hazardous unexploded ordnance (UXO) and non-hazardous shrapnel, range, and cultural debris. These improvements in classification performance have largely been achieved through advances in sensor design and signal processing methods developed in SERDP-sponsored research projects. In particular, geophysical surveys using instrumentation that maximize the amount of target information contained in the measured data are being combined with newly developed robust methods for extracting target parameters at the ESTCP live site demonstrations. Although classification results have been encouraging, these demonstrations have also shown that 100% recovery of all targets of interest (TOI) is not always possible without digging a significant number of non-TOI. In addition, there are still numerous instances where analysts have stopped digging before all TOI have been identified.
As the live site program has progressed to increasingly difficult sites, researchers have typically encountered shortcomings in the underlying UXO classification methodologies and have identified either enhancements to existing algorithms or completely new techniques and algorithms which improve classification performance. The goal of this project was to address these shortcomings and develop a robust and effective suite of processing and classification methodologies with wide applicability. The industry will then be in a position to tackle the full-spectrum of classification challenges encountered at the diverse UXO-contaminated sites spread across the country. Specific project objectives were to:
In this project, the researchers focused on the following five broad UXO classification topics:
The researchers extended and tested techniques that improve confidence in classifying buried items as either munitions and explosives of concern (MEC) or non-hazardous clutter. The researchers used, as a starting point, the techniques developed in SERDP project MR-1637, SERDP project MR-1573, and SERDP project MR-1629 that have the most promise for furthering the ability to classify MEC. These techniques were evaluated using data from the ESTCP live site demonstration program.
As increasingly difficult sites were addressed, the researchers encountered new, currently unforeseen challenges in classification that necessitated further development and fine-tuning of the techniques and algorithms researchers have developed. Testing and evaluation of these key techniques using live site data ensured that these techniques can be transitioned from novel concepts to robust components of practical solutions for the wide variety and ever expanding set of UXO classification challenges.
This research project considered several topics related to UXO data processing. Techniques were developed that improved various aspects of dynamic data processing. Filtering, sensor positioning, anomaly picking, and Informed Source Selection (ISS) methods were developed and tested. A range of multi-source inversion strategies for data collected at cluttered sites were tested. These approaches include techniques that optimize the data used for inversion, imaging approaches, multi-objective function inversions that penalizes overly complex models (e.g., redundant dipole sources), and global search optimization algorithms. The project team developed a Sequential Experiment Design (SED) method to improve in-field determination of instrument placement for data reacquired during cued surveys in a cluttered environment. To deal with positioning errors associated with multi-sounding inversions, they developed a method of inverting data called the Independent Model Location Inversion (IMLI).
Various practical approaches for mitigating the effects of magnetic soil were studied. The project team tested using a soil model that exploits relative transmitter/receiver pair amplitudes as well as the characteristic magnetic soil decays to improve processing performance. They provided examples of how scaling the background measurement should be considered at sites with magnetic soil noise to improve polarizability fits, and demonstrated how multi-source inversions can help the accuracy of polarizabilities by having sources represent background soil signal that was not sufficiently removed from the data through background correction. Methods of suppressing the soil effects that do not require subtracting a background measurement were investigated by using combinations or components of the data that are less sensitive to the magnetic geology.
Methods for automated review of inversion results and for quality control were developed. The project team looked at data based and model-based metrics to help eliminate spurious models, and for identifying potential TOI when that target’s polarizability is not in the site ordnance library. They developed a pair of classification approaches that does not require a priori information of target types and compared these techniques’ performance in finding novel targets when using a complete library.
Improvements in the classification of obscured items as either UXO or non-hazardous clutter has the potential to significantly decrease clearance costs. The main outcome of the project was a suite of processing and interpretation techniques that improve the efficacy and reliability of UXO classification. These methodologies are ready for incorporation into appropriate geophysical processing software.