Most pollutant gases can be detected by passive Fourier transform infrared(FTIR) remote sensing technique, and brightness temperature technique can implement the extraction and identification of target feature without the premise of background information. There usually exist radiation of background objects, radiance of atmospheric composition, and noise from the instrument itself in field measurements. When the target signal is weaker than these signals, the target feature can’t be extracted directly from the measured spectra by the brightness temperature technique. Accordingly, an extraction method of spectra feature from measured spectra based on non-negative matrix factorization(NMF) is proposed. It can get the target feature through the analysis of the brightness spectra on the basis of passive Fourier transform infrared remote sensing. The experiment is conducted with SO2 as the target gas. The method is verified with the measured spectra. Results show that the feature of SO2 can be extracted even if the target signal is weak. The validity of the method is proved.