The fluorescent spectrometry is a common method to detect the concentration of SO2 in the atmospheric monitoring. The detection system adopting double light paths can eliminate the noise jamming from the light source and light path. However, background noise produced by photoelectric converting device under the laser irradiation will also affect the accuracy of quantitative analysis. Empirical Mode Decomposition (EMD) filtering algorithm was used to reduce various kinds of noise existing in the detection, which could retain the useful original signal and reduce the noise effectively. The simulation results show that for the sulfur dioxide concentration detection system, using EMD de-noising, the Signal Noise Ratio (SNR) increases to 204.273 6, and the Mmean Squared Error (MSE) is 0.007 0. Compared with the wavelet de-noising method, the effect of EMD detection is much better. Finally, the signal processed with the two signal methods were applied to the gas detection system. From the experimental data of the linear relationship, it can be concluded that the EMD method applied to the proposed concentration detection system is feasible.