【摘要】 [目的]获得精度高、鲁棒性强的玉米近红外光谱淀粉组分检测模型。[方法]用一阶导数和Savitzky-Golay平滑对玉米1 300～2 298 nm近红外光谱进行预处理,而后分别以RS(random sampling)、KS(Kennard Stone)、Duplex、SPXY(sample set partitioning based onjoint x-y distance)方法选取最佳校正集样本集合,最后分别用PLS(Partial Least Squares)、iPLS(interval PLS)和siPLS(synergy intervalPLS)方法建立校正模型。[结果]采用SPXY方法选取有代表性的校正集合样本,以siPLS方法所建立的近红外光谱玉米淀粉组分校正模型最优,校正样本集合中r为0.991 7,RMSECV为0.107 3,预测样本集合中r达到了0.994 4,RMSEP为0.081 4。[结论]SPXY-siPLS方法建立的近红外光谱玉米淀粉组分校正模型,不但可以减小参与建模的数据规模,而且缩短了运算时间,预测能力和精度也均得到提高。更多还原

【Abstract】 [Objective] To obtain the testing model of starch content of near infrared spectrum in corn with high accuracy and strong robustness.[Method] The corn near Infrared spectra whose wavelength range was 1 300-2 298 nm was preprocessed by first derivative and Savitzky-Golay smoothing.The preprocessed spectra was selected for calibration set respectively adopting RS(random sampling),KS(Kennard Stone),Duplex and SPXY(sample set partitioning based on joint x-y distance) method.After the spectra being selected for calibration set,PLS(Partial Least Squares),iPLS(interval PLS) and siPLS(synergy interval PLS) were respectively used to set up the calibration model base on each calibration set of RS,KS,Duplex and SPXY method.[Result] The NIR calibration model of corn starch content established by SPXY-siPLS method was optimal,r of calibration set was 0.991 7,RMSECV was 0.107 3,r of prediction set was 0.994 4,RMSEP was 0.081 4.[Conclusion] The NIR calibration model of corn starch content established by SPXY-siPLS method could not only decrease the variable numbers of modeling and shorten the operation time,but also improve the prediction ability and precision.更多还原