A pre-process of feature extraction and classification approach based on spectrum edges matching is proposed to analyze the complicated nonlinear fluorescence spectra emitted by the interaction between femto-second (fs) laser and the impurities in air. The spectra data is denoised and compressed from 3979 points to 664 points using wavelet (WT) transform. By similarity analysis we create the characteristic spectra of 3 kinds of gases and the weights for classification. A new method of classification is proposed based on the edges matching and the comparison with characteristic spectra. Compared with existing methods，our method can not only get 100% classification accuracy，but also gives the characteristic position and the matching degree. The analysis of the matching degree shows that our method works well at low concentrations and has a potential application of identifying gases of lower concentration.