Spectral quantitative model optimization by modified successive projection algorithm
1.College of Mechanical and Electronic Engineering, East China Jiaotong University, Nanchang, Jiangxi 330013, China; 2.Ningbo EntryExit Inspection and Quarantine Bureau, Ningbo, Zhejiang 315012, China
Abstract：The raw and wavelet transform(WT) spectra of wine and apple were extracted by modified successive projection algorithm (MSPA).The quantitative models of alcohol and soluble solid contents (SSC) were established by partial least squares (PLS). Root mean square error of prediction(RMSEP), correlation coefficient of prediction (Rv) and akaike information criterion（AIC）were used to evaluate models. The results show that the simplized analysis models of alcohol and SSC determinations can be realized by WT-MSPA-PLS method. For alcohol model, RMSEP， AIC and modeling variables are reduced form 0.178, 4 085.60 and 2 073 to 0.139, -1.06 and 34, and Rv is improved from 0.963 to 0.976. For SSC model, RMSEP， AIC and modeling variables are decreased form 0.565, 1 047.20 and 535 to 0.541, 57.43 and 41, and Rv is increased from 0.920 to 0.930.
2013, Vol. 34 
