近红外光谱技术应用于玉米单籽粒蛋白质含量检测分析的初步研究  

Application of near-infrared spectroscopy technology to analyze protein content in single kernel maize seed

作　　者：张新玉[1] 王颖杰[1] 刘若西[1] 申兵辉[2] 王皎月[1] 严衍禄[3] 康定明[1] 

机构地区：[1]中国农业大学农学院,北京100193 [2]中国农业大学理学院,北京100193 [3]中国农业大学信息与电气工程学院,北京100193

出　　处：《中国农业大学学报》2017年第5期25-31,共7页Journal of China Agricultural University

基　　金：基金项目 光栅型近红外分析仪及其共用模型开发与应用(2014YQ470377)

摘　　要：为探索应用近红外光谱技术检测玉米单籽粒蛋白质含量,本研究采用JDSU近红外光谱检测仪采集了205份不同基因型玉米材料的单籽粒光谱值,用常规化学法测定玉米单籽粒蛋白质含量化学值,以117个样本为建模集,拟合了玉米单籽粒近红外光谱仪扫描得到的光谱图与玉米单籽粒蛋白质含量化学值之间的相互关系,用88个样本作预测集,比较了偏最小二乘回归法(PLSR)和支持向量机回归法(SVR)2种预测模型的效果。结果表明,玉米单籽粒种子的蛋白质含量在样本中变异范围为3.48%~18.15%,平均值为10.17%。偏最小二乘回归法(PLSR)和支持向量机回归法(SVR)所建的模型预测效果基本相同,其决定系数(R2)分别为0.99和0.99,校正标准差(SEC)分别为0.32和0.32,预测标准差(SEP)分别为0.46和0.46,相对预测标准差(RSEP)分别为4.61和4.60,RPD分别为6.106和6.111。上述参数表明PLSR和SVR所建立的模型预测效果都比较好,预测值基本接近参比值,便携式JDSU近红外光谱检测仪可以应用于定量分析玉米单籽粒蛋白质含量。The quality detection and classification identification of maize seed require to be accurately and quickly. The nutrient composition detection and identification method of maize seed currently are not only time consuming, but also inconvenient. Near-infrared spectroscopy can analyze samples in a short time without destroying samples. This study was focused on establishing a near infrared spectroscopy （NIR） technology to measure the protein content with a single maize seed, which was based on computer technology and chemometrics technology. A total 205 samples were exploited to construct a model, of which 117 samples were included in model dataset and 88 samples were for predicting dataset. A spectrum was obtained of each sample after diffused reflecting with a JDSU near infrared instrument. The chemical reference value of protein content of a single maize seed was analyzed. The protein content of a single maize seed sample with NIR was ranged from 3.48% to 18. 15%, and on the average, was 10. 17%. The simulating model between the chemical values and the spectrograph scanned by a JDSU, two modeling methods of Partial-Least-Squares Regression （PLSR） and Support Vector Machine （SVM） regression were adopted. The results showed that the effect of the two models built with PLSR method and SVM regression methods was almost same and the determination coefficient （R2） was 0.99 and 0.99, respectively. In addition, the standard deviation corrections （SEC） was 0.32 and 0.32, the standard deviation （SEP） was 0.46 and 0.46, the predictive relative standard deviation （RSEP） was 4.61 and 4.60,RPD was 6. 106 and 6. 111, respectively. It was demonstrated that the predicting model built by PLSR and SVR was better because the reference value is closer to the predicted value. And the analysis method to analyze the protein content of single maize seed via NIR JDSU was proved feasible.

关 键 词：玉米单籽粒 蛋白质含量 近红外 定量分析 

分 类 号：S513[农业科学—作物学]