Qualitative spectrum recognition algorithm is the basis of application of passive Fourier transform infrared spectroscopy (FTIR) in monitoring and early warning of toxic and hazardous gases. Three different methods, SFS, LASSO and Elastic Net, are used to perform preliminary selection of gas component variables in the FTIR spectrum by combining with the generalized cross-validation criteria. Then, the cyclic iterative CLS method is used to perform cyclic elimination of the components with concentration less than 0 in the preliminary screened variables. Finally, the direction proportion of the variables in the measurement vector is used to further select from the selected variables to get the target gas composition. In order to verify the performance of each recognition algorithm, the simulation experiments of CH4 and SF6 field emission are carried out respectively. The experimental results show that the established recognition algorithm can identify the target components quickly and effectively, the recognition response time is second level, the recognition accuracy is as high as 99%, and it can also accurately identify the interference component H2O. It is shown that the proposed algorithm provides a method basis for the application of passive FTIR technology in emergency monitoring and warning of dangerous gas leakage.