A spectral-spatial approach for detection of single-point natural gas leakage using hyperspectral imaging

Recent studies have shown that underground natural gas storage leaks can be indirectly detected through the spectral changes of surface vegetation. However, due to the phenomenon of different samples demonstrating the same spectrum, using a spectral-based approach may result in misdetection. Vegetation stressed by natural gas leakage has unique spatial patterns. Therefore, a field experiment of natural gas leakage vegetation stress was carried out. Hyperspectral images of bean, corn crops, and grasslands were obtained, which led to a proposed new spectral-spatial based methodology to detect natural gas leaks and areas of vegetation stress. First, the vegetation indices and the color index were extracted, then respectively segmented using the Otsu and the proposed threshold segmentation methods. Next, the shape parameters of the posture ratio and rectangularity of the segmented objects were used to construct a circular detection model. The accuracies of the detection results based on the vegetation indices and color index were 53 % and 56 %, respectively. Finally, based on the concentric ring spatial distribution pattern of the stress zones, the two types of detection results were fused using the linearly weighted fusion method, after which all the leakage points were accurately detected and localized, without any false alarms.