Identification and classification is the main purpose of hyperspectral imaging remote sensing pollutant gases, then the spatial distribution of the pollutant gases is obtained, as well as the location of the pollutant source. In practical applications, target spectrum is superimposed on intense background radiation, in addition, the spectra measured in open path comprise atmosphere interferences spectra which restrict identification and classification for target spectra. On the basis of linear model, orthogonal subspace projection method is used to effectively suppress the background and interferences′ information, and the subspace detector, based on gerneralized likelihood ratio test principle, is used to classify all pixels one by one. The field experiment is performed with ammonia as target gas, the data cube comes from scanning imaging Fourier transform infrared spectroscopy (FTIR) spectrometer, and the subspace vectors come from singular value decomposition (SVD). The recognition results for all pixels by subspace detector are superior to spectral angel mapper (SAM) algorithm.