点击次数：

论文类型：期刊论文

发表时间：2013-06-01

发表刊物：OPTICS AND LASER TECHNOLOGY

收录刊物：SCIE、EI、Scopus

卷号：48

页面范围：580-588

ISSN号：0030-3992

关键字：Photoacoustic spectroscopy; Feature extraction; Overcomplete ICA bases

摘要：In near-infrared band, macro representations of rovibrational-transition absorptions of some low molecule mass gases are "spiky" transient lines. The pressure broadening at ambient temperature and pressure make lineshape profiles at adjacent wavelength produce mutual projection or superposition, and it makes the "spiky" absorption features just be hidden in observable spectra. A blind source separation (BSS) model can extract features which hide within sampled spectra data, if a mixed gas phase photoacoustic signal is regarded as the weighted sum of feature absorptions by the spacing between adjacent feature wavelengths. A BSS model based on overcomplete ICA basis is proposed, and a weight truncation equation for transforming ICA bases based on the five-point sampling method is created so that a FastICA algorithm based on a negentropy approximation can be used to extract feature components. The resulting validity test using data from a real-time experiment of gas detection showed that the method detection limit can be improved from 16 ppb to 10 ppb with improved accuracy and signal-to-noise ratio. Crown Copyright (C) 2012 Published by Elsevier Ltd. All rights reserved.