玉米子粒颜色对近红外模型预测其淀粉含量的影响

以我国黑龙江、吉林、辽宁、内蒙古、山西、广西等省区368份玉米样品为材料,分别采用近红外透射光谱(NITS)和实验室常规分析(LAB)测定其淀粉含量,通过校正样品集建立样品吸收光谱与化学成分间的关系模型,校正并优化原有神经网络模型。经检验,校正模型Ⅰ(黄色样品)、校正模型Ⅱ(橙红色样品)和校正模型Ⅲ(总样品)的预测平均残差(Bias)依次为0.15、0.05和1.04;均方误差(RMSE)依次为1.06、1.07和1.12。预测效果相比较,模型Ⅰ和模型Ⅱ预测值与常规分析值的残差和均方误差均低于模型Ⅲ,因此对玉米整子粒淀粉近红外分析,按样品外观颜色分类定标(校正)更有利于提高模型的预测性能。

Effect of Kernel Color on Corn Starch Content by Near-infrared Transmittance Spectroscopy

Three hundred and sixty-eight corn variety samples were collected from Heilongjiang, Jilin, Liaoning, Neimeng, Shanxi and Guangxi Provinces in China. Starch content of corn varieties was analyzed by near-infrared transmittance spectroscopy and chemistry method. Calibration model was established with data from chemical analysis and absorbed spectrum of calibration samples to improve the accuracy of prediction results. The Bias values were achieved by calibration as 0.15 of model I(yellow-color samples), 0.05 of model II(orange-color samples) and 1.04 of model III (total samples). The RMSE values were 1.06, 1.07 and 1.12, respectively. Model prediction ability comparing showed that Bias value and RMSE value of model I and model II are all lower than the values of model III. It is useful to make the prediction model more practicable by calibrating separately for starch content analysis of whole kernel corn samples.