This project will demonstrate and disseminate physics-based and machine learning tools for delineation of Saturated Response Areas (SRA) in electromagnetic data acquired for Advanced Geophysical Classification (AGC). SRA are defined as regions where AGC cannot reliably identify targets of interest in the ground due to spatially extended, elevated signal that precludes estimation of dipole parameters required for classification. This elevated signal can be due to subsurface metallic infrastructure, a high concentration of metallic clutter in the ground, or a magnetic soil response. 

Data-based SRA delineation uses a gridded image of the data to identify spatially extended areas of elevated response, while model-based delineation relies on a map of anomaly (or source) density to identify SRA. Both methods require selection of thresholds (e.g. data amplitudes or source amplitudes), but there is currently no standard procedure for making these decisions.

This project will demonstrate a synthetic seeding procedure to objectively choose site-specific SRA decision points. Synthetic seeding adds modelled data for a target of interest with assumed extrinsic (location, depth, orientation) and intrinsic (polarizabilities) parameters to the observed AGC sensor data. Rather than assuming a simple noise model for simulations, this modeling approach uses the observed data to understand how noise, background response, anomaly density, and data coverage will affect classification. These simulations will allow the project team to tie SRA decision points more directly with the feasibility of classification under site-specific conditions. In addition, the project team will demonstrate how machine learning with convolutional neural networks can assist in the SRA selection process via identification of locations for synthetic seeding, efficient source density estimation, and prediction of the probability of correct classification given synthetically seeded data.

This project will benefit Department of Defense (DoD) and the AGC users by demonstrating recommended procedures for selecting SRA decision points—addressing a critical data useability need identified by industry stakeholders. Expected benefits will include a standardized process across the industry that increases stakeholder confidence in AGC.  

Reduced costs may be realized for industry and the DoD’s Military Munitions Response Program by ensuring that AGC is only applied in areas where it is reliable and minimizing the potential for rework. The demonstrated technology will be applicable across all munitions response sites, both marine and terrestrial, where dynamic sensor data is used to classify detected targets. (Anticipated Project Completion - 2027)