论文部分内容阅读
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。
The 368 corn samples from Heilongjiang, Jilin, Liaoning, Inner Mongolia, Shanxi and Guangxi provinces were used as materials to determine the starch content by near infra - red transmission spectroscopy (NITS) and laboratory routine analysis (LAB) The model of the relationship between the absorption spectrum of the sample and the chemical composition is corrected and the original neural network model is calibrated and optimized. After testing, the predicted average residuals (Bias) of the calibration model I (yellow sample), the calibration model II (orange red sample) and the calibration model III (total sample) were 0.15,0.05 and 1.04, respectively; mean square error (RMSE) 1.06, 1.07 and 1.12. Compared with the predicted results, the residuals and mean square errors of predicted value of model Ⅰ and model Ⅱ are lower than those of model Ⅲ. Therefore, for the analysis of neat grain near-infrared spectrum of corn, Help to improve the predictive performance of the model.