Doctoral Dissertations
Keywords and Phrases
Drones; Hyperspectral imaging; Machine Learning; Multimodal sensors; Pipeline leak; Remote sensing
Abstract
"This study explored the feasibility of using multimodal remote sensors in the early detection of natural gas leaks from underground pipelines, as vegetation situated above pipelines can discern microbial changes in soil when affected by gas and exhibit physiological stress symptoms on leaves. Three sensors (RBG, thermal, and hyperspectral cameras) were utilized to monitor vegetation stress as an indicator of gas leaks. A laboratory experiment was conducted to evaluate the workability of vegetation for gas leak detection. Regular hyperspectral imagery was collected from test vegetations to identify gas stress and distinguish it from other environmental stressors such as salinity impact, heavy metal contamination, and drought exposure. This study compared the sensitivity of different spectral zones (VNIR, SWIR, and VNIR+SWIR) in gas stress detection. It was found that vegetations can manifest gas leaks in hyperspectral images before any visible symptom appears on leaves. The first-order derivative of VNIR spectra yielded the highest detection accuracy in all scenarios. Drought exposure and salinity impact caused the least and most distractions, respectively, to gas detection from vegetation stress. Underground gas leaks were also simulated in an open field test and detected from drone-based imaging. Vegetation indicators (VIs) and deep neural networks (DNN) were proposed to extract and/or quantify features from images and compared for their efficiency and effectiveness in gas detection. Both VIs and DNN successfully detected natural gas leaks from hyperspectral imaging. DNN identified the gas leak after a 3-week treatment with an accuracy of 98.4%, which pinpoints a reduction in reflectance in near-infrared (NIR) spectrum coupled with an increase in 420-640 nm"-- Abstract, p. iv
Advisor(s)
Chen, Genda
Committee Member(s)
Wu, Chenglin
Yan, Guirong Grace
Ma, Hongyan
Gong, Weibing
Department(s)
Civil, Architectural and Environmental Engineering
Degree Name
Ph. D. in Civil Engineering
Publisher
Missouri University of Science and Technology
Publication Date
Spring 2025
Pagination
xvi, 170 pages
Note about bibliography
Includes_bibliographical_references_(pages 64, 114, 149and 160-167)
Rights
©2024 Pengfei Ma , All Rights Reserved
Document Type
Dissertation - Open Access
File Type
text
Language
English
Thesis Number
T 12464
Recommended Citation
Ma, Pengfei, "Drone-Based Multimodal Sensing On Vegetations And Data Analytics For Early Detection Of Gas Leakage From Underground Pipelines" (2025). Doctoral Dissertations. 3362.
https://scholarsmine.mst.edu/doctoral_dissertations/3362
