The Recognition of Nonlinear Fluorescence Spectra Based on a Support Vector Machine Networks

A support vector machine (SVM) network is applied to recognize the nonlinear fluorescence spectra of atmospheric impurity gases. As the number of spectrum channels is quite large, a wavelet transform (WT) is firstly adopted to remove the noises and to reduce the dimension of the data, afterward a principal component analysis (PCA) is used to extract the feature information. As result, the dimension of data is compressed from 3979 to 514 (WT) and finally to 9 (PCA), while the features of original nonlinear fluorescence spectra are remained. The results show that the correct recognition rate for the training samples and the testing ones has both reached 100%. Accordingly, the proposed method is efficient in inspecting of the atmospheric impurity gases in real time