饲料玉米品质分析的化学计量学方法

Chemometrics Method for Quality Analysis of Feed Corn

以饲料玉米为研究对象,采用化学计量学方法,利用全波段光谱数据建立玉米粗蛋白预测的简单快速精准预测模型。结果表明:原始光谱经去趋势算法预处理后,Rank-KS算法选择校正集和预测集,使用偏最小二乘(Partial least square,PLS)方法进行建模,校正集和预测集的相关系数分别为0.9915和0.9813,校正集和预测集的均方根误差分别为0.0634和0.1138。预测集的相对分析误差RPD为5.02,大于评估阈值3.0。所建模型精度和稳定性较为理想,可满足在线生成检测的要求。

Feed corn taken as the research object,stoichiometry was used to establish a simple and fast and accurate prediction model for maize crude protein prediction using whole-band spectral data.The results show that:After the original spectrum was preprocessed by detrending algorithm,the Rank-KS algorithm selected the correction set and prediction set,and used Partial least square(PLS)method for modeling.The correlation coefficients of the correction set and prediction set were 0.9915 and 0.9813,respectively.The root mean square errors of the correction set and the prediction set are 0.0634 and 0.1138,respectively.The relative analysis error(RPD)of the prediction set was 5.02,which was larger than the evaluation threshold of 3.0.The precision and stability of the model are satisfactory and can meet the requirements of on-line generation and detection.