Automatic Recognition of Overlapped Spectral Peaks by Combined Symmetric Zero-Area Conversion and L-M Fitting
BI Yun-feng1,2, LI Ying2, DU Zeng-feng2, CHEN Liang2, ZHENG Rong-er2*
1. College of Mechanicals, Electrical & Information Engineering, Shandong University, Weihai, Weihai 264209, China 2. Optics and Optoelectronics Laboratory, Ocean University of China, Qingdao 266100, China
Abstract：The peaks’ overlapping often exists in Raman spectroscopy analysis, because of the low spectral resolution of the spectrograph and the complex sample components. The overlapped peaks lead to the errors in peak parameters extraction easily, and at last lead to the analysis error of sample components, which increases the difficulty in automatic analysis of field spectra. The identification of overlapped peaks is the key difficulty of in-situ spectra analysis. To solve this problem, an automatic method of identifying the overlapped peaks was established basing on an analysis model with multiple Gaussian shape peaks. The peak number and the initial parameters(the peak position, peak height, and width) were obtained by symmetric zero-area transformation firstly, and then the parameters were optimized by Levenberg-Marquardt fitting method eventually. Some algorithm experiments were executed to test the method respectively by simulated data and Raman spectra data, and the former showed that the symmetric zero-area transformation method can extract the initial peak parameters with high accuracy, and then converges fast, and is adaptive to signal with wide dynamic range of SNR , but has false and omissive peaks to low SNR signal. The research results show that the automatic method of identifying the overlapped peaks with symmetric zero-area transformation combined with L-M fitting has a certain practical value.
Key words：Recognition of overlapped spectral peaks;Levenberg-Marquardt fitting;The symmetric zero-area conversion
