Abstract：A Soft Thresholding Wavelet-based Radial Basis Function Neural network (STWRBFN) method was developed to perform simultaneous quantitative analysis of multicomponent mixtures. The quality of noise removal and regression was improved by combining wavelet soft thresholding with radial basis function neural network. Through optimization, the wavelet function, wavelet decomposition level (L), thresholding method and spread parameter σ of RBFN were selected. Two-programs, i.e. PSTWRBFN and PRBFN, were designed to perform STWRBFN and RBFN calculations. Experimental results showed the STWRBFN method to be successful and better than RBFN. Comparing with classical multivariate linear regression, both the methods were more powerful.
Key words：Wavelet soft thresholding;Radial basis function neural network;Simultaneous multicomponent analysis
[1] 梁逸曾, 俞汝勤．分析化学手册：10，化学计量学. 北京：化学工业出版社，2001. 95. [2] Gao Ling, Ren Shouxin．Chemometrics and Intelligent Laboratory Systems, 1999, 45: 87. [3] Zupan J, Gasteiger J. Neural Network for Chemistry. New York：VCH publisher, 1993. [4] Alsberg B K, Woodward A M, Kell D B. Chemometrics and Intelligent Laboratory Systems, 1997, 37: 215. [5] J Orr M I. Neural Comput., 1995, 7: 606. [6] Ren Shouxin, Gao Ling．Talanta, 2000, 51: 1163. [7] Daubechies I. Commun. Pure Appl. Math., 1989, XLI: 909. [8] Daubechies I. SIAM J. Math Anal., 1993, 24(2): 499. [9] Cohen A, Daubechies I. SIAM J. Math Anal., 1993, 24(2): 520. [10] Donoho D L. IEEE Transaction of Information Theory, 1995, 41(3): 613. [11] Hilton M L, Ogden R T. IEEE Transactions on Signal Processing, 1997, 45: 496. [12] Mittermayr C R, Frischenschlager H, Rosenberg E. Fresenius J. Anal. Chem., 1997, 358: 456. [13] Stein C. Ann. Stat., 1981, 10: 1135. [14] Holohninger J. Chem. Inf. Comput. Sci., 1993, 33: 736. [15] CHEN Li-li, JIN Ji-hong, QIN Sun-wei(陈莉莉，金继红，秦孙巍). Spectroscopy and Spectral Analysis(光谱学与光谱分析)，2003，23(3)：597.
