摘要
以 2 0 0 1和 2 0 0 2年度 5 83份小麦样品为材料 ,用近红外透射光谱仪 (NITS)对小麦籽粒硬度进行分析 ,比较了偏最小二乘法和多元线性回归两种算法和未经导数处理、一阶导数处理、二阶导数处理 3种光谱变量转换方式的分析结果。表明 ,两种算法中偏最小二乘法优于多元线性回归算法 ,3种处理方式中一阶导数处理效果最好 ,其定标集和预测集决定系数明显高于其他两种处理方式 ,而标准误差低于其他两种方式。经一阶导数处理后采用偏最小二乘法建立的定标模型时 ,可有效地对小麦籽粒进行硬度分级 ,其中 ,硬麦分级准确率为 90 % ,软麦分级准确率为 83% ,混合型分级准确率为 6 3%。
samples of wheat cultivars from major wheat production area collected in 2001 and 2002 seasons were used to measure kernel hardness by near infrared transmittance(NIT)spectroscopy. Two algorithms, i.e. partial least squares and multiple linear regression, and three variable transformations, i.e. log 1/T, first derivative of log 1/T and second derivative of log 1/T, were compared for hardness testing. Results showed that the partial least squares and the first derivative of log 1/T were preferable, and the classification accuracy was achieved 90% for hard type, 83% for soft type, and 63% for mixed type, respectively. Wheat kernel hardness test by NIT spectroscopy could be used for early generation selection in wheat breeding program and quick testing for wheat quality.
出处
《作物学报》
CAS
CSCD
北大核心
2004年第5期455-459,共5页
Acta Agronomica Sinica
基金
国家 8 63计划 ( 2 0 0 1AA2 410 3 1
2 0 0 2AA2 0 70 0 3 )
973重点发展研究规划 ( 2 0 0 2CB1113 0 0 )
948重大国际农业合作和国家自然科学基金 ( 3 0 2 60 0 61)资助。
关键词
近红外透射光谱技术
小麦
籽粒硬度
定标模型
品质
Near infrared transmittance spectroscopy (NITS)
Calibration model, Common wheat
Hardness