【作者】 彭建；

【导师】 张正茂；

【作者基本信息】 西北农林科技大学 ， 粮食油脂及植物蛋白工程， 2010， 硕士

【摘要】 小麦是世界上最重要的谷物资源之一,产量仅次于稻谷居第2位。小麦产量对世界经济有重要影响,在农业和工业中都需要对小麦的品质和品种进行分析和鉴别。本论文试验以关中地区小麦为材料,用德国Bruker公司的MPA近红外扫描仪进行籽粒的光谱扫描,同时对小麦籽粒的总淀粉含量、直链淀粉含量和蛋白含量进行化学检测,用OPUS 5.5、SAS 8.1、DPS 6.55分析软件对数据进行分析,建立小麦品质测定和品种鉴别的近红外模型,主要的研究结果如下:小麦籽粒品质分析的近红外模型均用偏最小二乘法(PLSR)建立。小麦籽粒淀粉近红外模型的最佳建模方法为:预处理方法选择一阶导数+减去一条直线,光谱范围选择7501.9 cm-1~5450 cm-1,主成分数确定为7;模型的交叉验证决定系数R2cv达到0.6902,交叉验证标准差RMSECV值为1.08,外部验证决定系数R2val为0.8151,外部验证标准差RMSEP达到0.878。小麦籽粒直链淀粉近红外模型的最佳建模方法为:预处理方法选择矢量归一化,光谱范围选择7501.9 cm-1~4597.6 cm-1,主成分数确定为8;模型的交叉验证决定系数R2cv达到0.8552,交叉验证标准差RMSECV值为0.638,外部验证决定系数R2val为0.8403,外部验证标准差RMSEP达到0.675。小麦籽粒蛋白质近红外模型的最佳建模方法为:预处理方法选择一阶导数+减去一条直线,光谱范围选择10502 cm-1~6098 cm-1,主成分数确定为9;模型的交叉验证决定系数R2cv达到0.9528,交叉验证标准差RMSECV值为0.402,外部验证决定系数R2val为0.9449,外部验证标准差RMSEP达到0.422。小麦品种鉴别模型的建立:依据光谱分析,选择光谱范围4000 cm-1~10500 cm-1,采用Savitzky-Golay平滑法,平滑点选择9,进行平滑处理,滤除噪声,再用矢量归一化处理。将处理后的光谱数据导入SAS中进行主成分分析。将得到的主成分作为新变量,分别用三种不同的模式识别方法(BP-人工神经网络、Fisher多类线性判别、Bayes多类逐步判别)分析,建立小麦品种鉴别模型,以模型的正确分类率来衡量其好坏。BP-人工神经网络分析方法选择6个主成分建立了一个6(输入层节点)—5(隐含层节点)—1(输出层节点)的三层ANN-BP模型,建立模型的校正集和验证集的正确分类率均达到100%;Fisher多类线性判别分析方法选择6个主成分分析,所建模型校正集的正确分类为96.8%,验证集的正确分类率达到100%;Bayes多类逐步判别分析方法选择3个主成分,所建模型的校正集和验证集的正确分类率均达到100%。比较分析,主成分分析结合Bayes多类逐步判别方法建立小麦品种鉴别模型效果较好。建立小麦籽粒品质和品种的近红外模型是可行的,所建模型均有较好的预测效果。更多还原

【Abstract】 Wheat is one of the word’s most important grain resource, the only yield in the 2nd. Wheat yield has important effect on world economy. The quality analysis and species identification are needed in agriculture and industry. This paper takes wheat in central Shaanxi area as samples, scan the wheat by MPA from German Bruker Corporation and then get the starch content,amylose content and protein content of whole wheat by chemistry test. With OPUS 5.5, SAS 8.1, DPS 6.55 analysis softwares to analyze data, establish near-infrared spectroscopy(NIRS) models of quality analysis and species identification. The main research results are as follows:The NIRS models of wheat quality were established by partial least square regression(PLSR). The optimal NIRS modeling method of wheat starch are that pretreatment method is FD+SLS, NIRS region is 7501.9 cm-1~5450 cm-1 and the number of principal component is 7. The model parameters are that R2cv value achieves 0.6902, RMSECV is 1.08, R2val is 0.8151 and RMSEP achieves 0.878. The optimal NIRS modeling method of wheat amylose are that pretreatment method is SNV, NIRS region is 7501.9 cm-1~4597.6 cm-1 and the number of principal component is 8. The model parameters are that R2cv value achieves 0.8552, RMSECV is 0.678, R2val is 0.8403 and RMSEP achieves 0.675. The optimal NIRS modeling method of wheat protein are that pretreatment method is FD+SLS, NIRS region is 10502 cm-1~6098 cm-1 and the number fo principal component is 9. The model parameters are that R2cv value achieves 0.9528, RMSECV is 0.402, R2val is 0.9449 and RMSEP achieves 0.422.Wheat cultivars identification model was established.based on spectral analysis, select NIRS region 4000 cm-1~10500 cm-1 and apply smoothing process(Savitzky-Golay smoothing method and smoothing point 9) to filter noise and SVN process. Then import the spectral data processed into SAS analysis software for PCA. The principal components computed by PCA was used as the new variables and analysized by three different methods of pattern recognition (BP-ANN, FLDA,BSDA) respectively, Wheat cultivars identification model was set up. The correct classification rate of identification model was used to measure its quality. With 6 principal components as input vector, and the nodes of input layer, pattern layer and output layer was 6, 5, 1, respectively, the ANN-BP model was build up well and the correct classification rates of calibration set and validation set are both 100%; Select 6 principal components to apply FLDA and the correct classification rates of calibration set and validation set are 96.8%, 100%, seprately; Select 3 principal components to apply BSDA and the correct classification rates of calibration set and validation set are both 100%. Through the comparion, the effect of the wheat cultivars identification model by principal component analysis compared with BSDA method is good.It is feasible to establish NIRS model of wheat grain quality and cultivars. The effect of all models is good.更多还原