采用反向区间偏最小二乘法和组合区间偏最小二乘法优化桃糖度可见/近红外光谱的信息区间组合, 在选择 的信息区间的基础上建立了一种线性组合权重PLS模型.对近红外光谱进行二阶导数处理、卷积平滑校正后, 发现 在区间分割数为15时筛选结果最优, BiPLS所选择的信息区间为742～770nm和 862～920nm, SiPLS所选择的信息 区间为742～770nm、832～860nm和892～920nm.直接组合信息区间BiPLS和SiPLS模型的RMSEP值分别为0.386 和0.308, 线性组合权重PLS模型的RMSEP值分别为0.351和0.364.结果说明在近红外定量分析中线性组合权重 模型的建立克服了复杂样品各信息区间对PLS建模贡献率不一样的问题.

Backward interval partial least squares (BiPLS) and synergy interval partial least squares (SiPLS) were pro- posed to search for an optimized combination of information spectral intervals about soluble solids content (SSC) from Vis/ NIR spectra of peach. A linear combination weight PLS model was developed on the basis of the selected information inter- vals. The spectra were preprocessed by second-order derivative and Savitzky-Golay smoothing. It is found that the selected result is the best when the interval number is 15. The information intervals selected by BiPLS are 742～770nm and 862～ 920nm, while those selected by SiPLS are 742～770nm, 832～860nm and 892～920nm. For BiPLS and SiPLS models of direct combination intervals, the root mean square error of prediction (RMSEP) are 0.386 and 0.308, respectively. And for the PLS models of linear combination weight, the RMSEP are 0.351 and 0.364, respectively. The results reveal that the proposed method overcomes the difficulties that the different information intervals for the complicated samples have dif- ferent contribution to PLS models. The metlod is very promising for vibrational spectroscopy and it gives much better predic- tion than the whole-spectrum PLS modeling.