Gas monitoring is closely related to our lives, and hydrogen as an ideal research model has received widespread attention. Raman spectroscopy has the advantages of non-destructive and non-contact measurements. One of the main problems in Raman measurement for gases is the weak Raman scattering. In some specific scenarios, the signal acquisition time is required to be short, so the obtained Raman spectrum has a low signal-to-noise ratio. As a newly developed signal processing method, the compressed sensing method can compress and sample the signal, shorten the sampling time, and reduce the noise and improve the signal-to-noise ratio to better realize the restoration and reconstruction of the original signal. This study used hydrogen and deuterium mixed gas as the measurement object. Two compressed sensing methods were used to analyze the Raman spectra with different sparse matrices: One of the sparse matrices using the Lorentz function to design atoms of the dictionary of OMP (Orthogonal Matching Pursuit) algorithm, and another sparse matrix using Fourier transform filtering to construct the orthogonal basis dictionary of OMP. Through the processing of simulation data and actual measurement data, we compare the effects of the two methods of compressed sensing analysis with wavelet soft threshold, hard wavelet threshold and SG (Sawitzky-Golay) filter processing, and peak intensity, signal-to-noise ratio, and root mean square error. It is demonstrated that using the Lorentz function to design the dictionary of the OMP algorithm can reduce noise for gas Raman spectra.