玉米单籽粒及单穗籽粒直链淀粉质量分数NIRS模型的建立与验证

The Calibration and Validation of NIRS Prediction Models for Amylose Mass Fraction of Single-kernel and Single-spike of Maize

籽粒直链淀粉质量分数的快速、无损伤测定是高直链淀粉玉米育种的关键环节。以196份高直链单籽粒和360份高直链单穗玉米籽粒为样本,分别利用碘染色法和一阶导数+标准正态变量变换(SNV)的光谱预处理法,构建单籽粒和单穗籽粒直链淀粉质量分数的NIRS分析模型,并通过分割建模样品化学值变异范围的方法建立2个单穗籽粒的NIRS子模型,以期提高对单穗籽粒样品的预测准确度。结果表明,所建立的4个模型的交叉验证标准偏差(RMSECV)分别为1.805、3.370、2.394、2.408,预测标准偏差(RMSEP)分别为2.017、3.205、2.369、2.596,各项决定系数(R2cal、R2cv、R2val)为0.626 1~0.897 0。表明,所建玉米单籽粒NIRS模型的预测准确度较高,可用于早代玉米单籽粒直链淀粉质量分数的鉴定;单穗NIRS子模型能够在一定程度上弥补单穗NIRS模型在预测准确度上的不足,将总模型与子模型配合使用能提高预测准确度。

Determination of amylose content rapidly and nondestructively is the key point of high-amylose maize breeding.In this study,a total of 196 high-amylose kernels and 360 high-amylose spikes with different amylose content were employed to build single-kernel and single-spike Near Infrared Spectroscopy（NIRS）models with assistance of iodine staining and spectra preprocessing method of First Derivative and Standard Normal Variable Transformation（SNV）.Meanwhile,to improve the predicting accuracy,training samples was divided by chemical value range of amylose content into two groups to build another two NIRS sub-models of single-spike.The root mean square error of cross validation（RMSECV）of 4 models was respectively 1.805,3.370,2.394,2.408,while coefficients of determination（R_（cal）~2,R_（cv）~2,R_（val）~2）ranged from 0.626 1 to 0.897 0.The results showed that the prediction accuracy of single-kernel NIRS models was high enough to determine amylose content of single-kernel of early generation;single-spike NIRS sub-models could make up the weakness on predicting accuracy of single-spike NIRS model,thus,the predicting accuracy could be improved with the coordinate application of the general model and sub-models.