玉米完整籽粒近红外品质分析模型的比较及改进

Comparison and Improvement of Quality Analytical Models for Whole Maize Kernels Using Near Infrared Spectroscopy

本文对国家玉米改良中心品质分析实验室采用普通玉米和高油玉米建立的完整籽粒近红外品质(蛋白质、油分、淀粉)分析模型(简称模型Ⅰ和模型Ⅱ)进行了比较分析,并对它们进行了进一步地改进。比较分析发现模型Ⅰ蛋白质预测的决定系数(R2),定标均方根差(RMSEE),外部验证决定系数(Rp2)和外部验证均方根差(RMSEP)与模型Ⅱ相差不大,模型Ⅰ蛋白含量范围8.15%～16.06%,比模型Ⅱ8.55%～14.7%稍宽;改进以后的模型Ⅲ蛋白质含量8.15%～16.06%,分布比模型Ⅰ和2要广,浓度梯度更均匀;模型Ⅰ油份含量范围2.85%～10.26%,适合测低油分籽粒,模型Ⅱ油份含量范围4.16%～10.12%,适合测高油分籽粒,这与其建模时分别使用普通玉米和高油玉米材料相吻合。模型Ⅲ油分(2.85%～10.26%)涵盖了模型Ⅰ的普通玉米和模型Ⅱ的高油玉米,适应性更广。模型Ⅰ高淀粉含量籽粒分布多,模型Ⅱ低淀粉含量籽粒分布多;模型Ⅲ综合了模型Ⅰ的高淀粉含量玉米和模型Ⅱ的低淀粉含量玉米,适应范围也得到加强。

The near infrared spectroscopy （NIRS） quality （ protein, oil, starch） analytical models for maize kernels （ model Ⅰ and model Ⅱ ）, which were established by using maize inbreed lines and hybrid varieties and by using offspring derived from hybrid variety of common maize and high oil maize in the analysis laboratory of National Maize Improvement Center of China, were investigated and compared ; and an improved model （ model Ⅲ ） was developed. Through the comparison it is found that protein model Ⅰ has little discrepancy with model Ⅱ in protein determination coefficient of estimation（ R^2 ）, determination coefficient of prediction（ Rp^2）, root mean square error of estimation（ RMSEE）, and in root mean square error of prediction （RMSEP） ; it is also found that model Ⅰ is wider in protein content adaptive range （8.15% - 16.06% ） than model Ⅱ （8.55% - 14.70% ）. The protein content adaptive range of the improved models （ model Ⅲ ） is more wider and the grads of concentration are more uniform. Oil model Ⅰ is applicable to predict the low oil kernel, and model Ⅱ is applicable to predict high oil kernel, as they were established by using high oil and low oil kernel respectively. The oil content range of model Ⅰ is slightly wider than model Ⅱ. The oil content of model Ⅲ covers high oil content of model Ⅱ and low oil content of model Ⅰ , and the adaptability is more extensive. High starch kernels distribute wide in starch model Ⅰ, while low starch kernels distribute wide in starch model Ⅱ. In starch content the model Ⅲ colligates high starch maize kernels of model Ⅰ and low starch maize kernels of model Ⅱ , and the adaptive range is enhanced.