Optional ZOOM LINK
Meeting ID: 919 5532 4108
Passcode: 289044

Abstract: Remote sensing of the subsurface is a critical need, including searching for targets such as buried infrastructure, unexploded ordnance, and even geological formations and structures, encompassing a humanitarian need as well as environmental and ecological ones. The reliable identification of the 34 million acres of land considered UXO-contaminated in Ukraine since 2022 or the coming overhaul of US drinking water pipelines, as examples, highlight the importance of not only having a reliable method to identify targets of interest in the subsurface, but also to go from data collection to decision-making in a streamlined and quick manner.

Much prior work has been done to develop non-invasive geophysical sensors to conduct surveys, such as magnetometers, ground-penetrating radar, & advanced electromagnetic induction (EMI) sensors, as well as data processing algorithms to detect, locate and identify targets. Most of these technologies are heavy and provide mainly target detection, or require multiple passes over the same targets. Typical data processing methodologies include analyzing an initial dynamic data collection to identify potential targets, then conducting static collections over those areas of interest, commonly applying library-matching techniques to classify targets. This impedes operation speeds and leads to higher costs for detection, mapping, and identification of subsurface targets.

This proposal encompasses hardware packaging and experimental testing, as well as numerical analysis, modeling, and algorithm development. The first of these (1) is the integration of a lightweight EMI sensor onto moving platforms. The second (2) is combining EMI and magnetometer sensors to improve deep target detection and identification. Integrating multiple detection modalities is intended to aid in detection of targets, by leveraging each individual sensor's strengths to overcome another’s drawbacks. The third (3) is the development of algorithms to detect and localize a magnetic dipole source using a closed-form analytical solution, and initial simulations of potential scenarios. The final topic (4) is applying machine learning algorithms to further improve subsurface and underwater target classification using the combined EMI and magnetometer sensors. Prior work indicating support vector machines (SVM) feasibility, as well as implementing onboard trained neural networks, will be focused on for real-time target classification.

Thesis Committee: Fridon Shubitidze (chair), Ben Barrowes, Laura Ray, George Cybenko, Lev Eppelbaum (Tel Aviv U, Israel)