As a clean, efficient low-carbon energy source, natural gas accounts for an increasing proportion of consumption. For underground gas pipelines, gas storage, and the like, natural gas leakage will occur due to factors such as pipeline corrosion, aging, natural disasters, underground faults, and the bad sealing of injection-wells. In terms of security, economic, environmental and other considerations, micro-leak detection of underground natural gas pipeline and gas storage is essential. In this paper, we use hyperspectral remote sensing to monitor surface vegetation changes, thus indirectly detecting natural gas micro-leakage points. The field controllable system is used to simulate the underground micro-leakage experiment. Winter wheat is used in this study, and a time series of 9 experiments of canopy spectral collecting were conducted. Spectral analysis was used to identify and exploit the spectral characteristics of stress wheat and thereby constructing an index recognition model. Firstly, the wheat canopy spectrum is subjected to the processing of singular value culling and smoothing, and then the continuous spectrum wavelet transform is performed on the canopy spectrum after continuum removal. Specifically, mother wavelet of Mexihat is selected. When the scale parameter is 32, the wavelet coefficients have fewer peaks and valleys, which can fit well with the original spectrum, and the peak and valley positions of wheat multi-phase data are relatively stable. The original spectrum of stress and healthy wheat was poorly separable, but separability of proposed model using the wavelet coefficients at 487, 550 and 770 nm was better among wheat samples, and had obvious diagnostic characteristics: (1) The wavelet coefficient of stress and healthy wheat having “absorption valley” at 487 nm, the wavelet coefficient value being negative, and the wavelet coefficient of healthy wheat being larger than that of stressed wheat; (2) the wavelet coefficient of stress and healthy wheat is 550 nm at 770 nm, where clear “reflection peak” can be observed, and the wavelet coefficient of stressed wheat is larger. In order to better highlight the differences of wavelet coefficients of stressed and healthy wheat, the index CWTmexh(CWTmexh=CW2770/(1-CW487)·CW550) was constructed for the identification of stress and healthy wheat. Compared with the index NDVI705, mNDVI705, ARI1, R440/R740, D725/D702 and J-M distance quantitative test, the results show that the CWTmexh has a better recognition performance on winter wheat under natural gas micro-leakage stress. The CWTmexh can stably distinguish between stress and health after 20 days of natural gas stress, and maintain the same performance in the whole growth period, while the indexes of NDVI705, mNDVI705, ARI1 and so on, can not accurately identify throughout the growth period. The CWTmexh index is superior to the other five indexes in terms of stability, universality and the ability to recognition. Therefore, it is feasible to indirectly identify natural gas micro-leakage points by monitoring surface vegetation using hyperspectral remote sensing. The results can provide theoretical basis and technical support for monitoring underground gas leakage points by satellite-borne hyperspectral remote sensing.