LAUSR

Abstract Spectral noise has direct influences on the optical electronic nose (e-nose) to predicted effect and the detection accuracy of the test gas. In this paper, a spectral denoising method based on least squares support vector machine (LSSVM) was proposed after comprehensive analysis of spectral characteristics including small sample, non-linear, local extreme points and so on. The optical e-nose sensing data of NO2, SO2, C6H6 and C7H8 collected by the system were denoised by LSSVM. The normalized correlation coefficient (NCC) between the deniosed spectrum and the standard spectrum in HITRAN database is over 98%. Compared with the results of moving window average filtering (MWA), Savitzky-Golay filtering (SG) and wavelet threshold filtering, it is found that the waveform obtained by LSSVM can retain the information such as relative extreme value and width of the spectrum peak. The validity and superiority of LSSVM were verified by the experiment.