摘要
为了实现田间快速、无损测定整粒小麦籽粒蛋白质含量，于2011-2015年，以收集的176个蛋白质含量有明显差异的冬小麦品种为研究对象，在开放环境下测定了不同品种小麦籽粒反射光谱和蛋白质含量。将预处理后的小麦籽粒光谱，在400-2400nm和780—2400nm波段范围内经过剔除无信息变量（UVE）与连续投影算法（SPA）获得蛋白质的特征波段，与小麦籽粒蛋白质含量进行交叉验证偏最小二乘回归分析，构建并优选小麦籽粒蛋白质含量的光谱最佳定量估测模型；比较不同预处理与不同特征波段筛选方法组合所得预测模型后发现，利用51点一阶导数处理后的近红外光谱经过UVE与SPA组合筛选的特征波段构建的模型对整粒小麦籽粒的蛋白质含量预测精度最高，可靠性好，该方法经济有效，预测均方根误差（RMSEP）和R2分别为0．4856和0．9653。可以实现田间无损、快速测定整粒小麦籽粒蛋白质含量。
In order to rapidly and nondestructively acquire grain protein content of wheat in the field, 176 cultivars of winter wheat with significant difference in protein content were selected to test the reflectance spectra and grain protein content in open environment. The pretreated grain reflectance spectra was used to obtain characteristic wave bands of protein content by uninformative variables elimination （UVE） in the range of 400-2400nm and 780-2400nm and by successive projections algorithm （SPA）. Using partial least squares regression analysis method cross validated the protein content of wheat and its characteristic band, the optimal quantitative estimation model of grain protein content were developed. Developed with the spectra pretreated methods and different feature selection methods, we found that the estimation models of wheat protein content ,which are developed by using the 31 point first derivative of the near infrared spectrum, and used UVE and SPA to screen protein content characteristic bands, have highest estimated precision and reliability for grain protein content of wheat. These models are economical and effective of which RMS error prediction （RMSEP） and R2 is 0.4856 and 0.9653 respectively, and can estimate grain protein content in wheat rapidly and nondestructively in the field.
出处
《现代科学仪器》
2016年第1期69-75,共7页
Modern Scientific Instruments
基金
国家科技支撑计划（2012BAH27R004）,国家自然科学基金（31460317）
关键词
小麦
籽粒蛋白质
开放环境
定量分析
反射光谱
模型构建
Wheat
Grain protein
Open environment
Quantitative analysis
Reflectance spectra
Model construction
