Data Science Ph.D. Qualifier | Noushin Khosravi Largani | Monday, Dec. 2nd @ 4pm, Olin 218

DATA SCIENCE

Ph.D. Qualifier Presentation

Noushin Khosravi Largani

Monday, December 2nd, 4 pm - 5 pm

Location: Olin Hall 218

Committee: 

Seyed Reza Zekavat, advisor

Oren Mangoubi, co-advisor1

Raha Moraffah, co-advisor 2

Title:  Soil Subsurface Channel Statistical Characterization for Intelligent GPR Waveform Design

Abstract: 

The advancement of drone-borne intelligent Ground Penetration Radar (GPR) is hinged upon accurate received signal feature extraction, which relies on the optimal design of the transmitted waveform. Optimal waveform design needs soil subsurface channel parameter statistics (CPS). In drone-borne GPR applications, the GPR should utilize a limited number of received signals to rapidly pick an optimal waveform. This rapid selection will be possible via Machine Learning (ML) models. This selection is a two-step ML process: (i) ML that maps GPR signals into soil CPS, and (ii) ML that adopts an optimal waveform for a given CPS. This work aims to investigate the first step, i.e., ML that maps GPR signals into soil CPS. There are different types of soil texture, moisture levels and composition which impacts the CPS and the received signal. Gathering numerous real data for different types of soil texture, moisture and composition to create an ML model is expensive. This work aims to use gprMax software to generate synthetic labels that map GPR received signal into soil subsurface CPS. The gprMax is a strong emulation tool for synthetic GPR signal generation. We study the details of proper gprMax design that generates data consistent with realistic soil scenarios. The gprMax is computationally expensive. We also assess adoption of the parameters that maintain a tradeoff across complexity and accuracy. Next, we use gprMax emulations to derive the CPS for different soil classes. The results of this study will be used to develop ML models that map received signals to the soil subsurface CPS.