Natural Gas Induced Vegetation Stress Identification and Discrimination from Hyperspectral Imaging for Pipeline Leakage Detection

Abstract

Remote sensing detection of natural gas leaks remains challenging when using ground vegetation stress to detect underground pipeline leaks. Other natural stressors may co-present and complicate gas leak detection. This study explores the feasibility of identifying and distinguishing gas-induced stress from other natural stresses by analyzing the hyperspectral reflectance of vegetation. The effectiveness of this discrimination is assessed across three distinct spectral ranges (VNIR, SWIR, and Full spectra). Greenhouse experiments subjected three plant species to controlled environmental stressors, including gas leakage, salinity impact, heavy-metal contamination, and drought exposure. Spectral curves obtained from the experiments underwent preprocessing techniques such as standard normal variate, first-order derivative, and second-order derivative. Principal component analysis was then employed to reduce dimensionality in the spectral feature space, facilitating input for linear/quadratic discriminant analysis (LDA/QDA) to identify and discriminate gas leaks. Results demonstrate an average accuracy of 80% in identifying gas-stressed plants from unstressed ones using LDA. Gas leakage can be discriminated from scenarios involving a single distracting stressor with an accuracy ranging from 76.4% to 84.6%, with drought treatment proving the most successful. Notably, first-order derivative processing of VNIR spectra yields the highest accuracy in gas leakage detection.