【作者】 鹿保鑫；

【导师】 张东杰；

【作者基本信息】 黑龙江八一农垦大学 ， 农业机械化工程， 2017， 博士

【摘要】 黑龙江省是我国目前大豆的主产区,大豆产量占全国30%以上,但存在着产地溯源技术不完善、品牌保护技术缺失等问题。大豆生产过程中的耕整地、播种施肥、田间管理和收获环节的全程机械化,是实现产地溯源规范化的必要条件。为解决我国大豆溯源技术总体水平较低的问题,本文以黑龙江省全程机械化作业的黑龙江农垦北安管理局和嫩江中储粮北方农业开发有限公司两个产区的大豆及土壤为研究对象,通过对两个产区2015年和2016年的大豆有机成分及矿物元素含量进行测定,利用方差分析、主成分分析、聚类分析、判别分析等方法,筛选出以3种有机成分和7种矿物元素作为溯源特征指标,建立了大豆产地的判别模型,结合线性判别分析的维数规约和支持向量机算法,开发了基于支持向量机算法的产地判别系统。研究结果及结论如下:(1)通过对两个大豆产区试验田2015年和2016年随机采集的97份大豆样品中的蛋白质、脂肪、灰分及可溶性糖含量进行测定,利用方差分析、主成分分析、聚类分析和线性判别分析法,探寻表征大豆产地溯源有机成分的特征指标。结果表明,利用有机成分建立的判别模型对产地的整体正确判别率及整体交叉检验判别正确率均达到了86.0%。(2)以两个产区试验田2015年和2016年的113份大豆及对应土壤样品为研究对象,测定了大豆及对应土壤中52种矿物元素,并通过方差分析、主成分分析、聚类分析和线性判别分析法,探寻表征大豆及对应土壤中产地溯源矿物元素的特征指标。结果表明,不同产地来源的大豆样品中矿物元素的含量具有显著差异;利用Mg、Mn、Sr、La、Gd、Tb、Hf、Ti等8种特征的矿物元素建立的判别模型对两个地区大豆产地判别正确率达到93.2%。(3)以2015-2016年采集的两个产区112份样本作为研究对象,以2014年随机采取的56份样本做为验证对象,筛选出Mn、As、Sr、La、Nd、Tb、Hf等7种矿物元素和蛋白质、脂肪、可溶性糖等3种有机成分作为产地溯源的特征指标,并建立了Fisher产地判别模型,结合线性判别分析的维数规约和支持向量机算法进行大豆产地判别,正确判别率为94.6%,优于线性判别模型的92.9%。(4)采用EF网页框架搭建了基于MVC模式的系统,建立了大豆矿物元素及有机成分数据库,设计开发了基于支持向量机算法的大豆产地判别系统,应用该系统对大豆产地进行判别,正确判别率为97.5%,优于线性判别模型的90%。建立的大豆产地判别追溯系统具有较高的判别率,为将来开展大豆产地溯源奠定理论基础。更多还原

【Abstract】 Heilongjiang province is the main soybean production region in China,which accounts for more than 30% of the national soybean production amount.However,problems such as the imperfect Origin Traceability Technology and the lack of brand protection technology are still exist.In the process of soybean production,whole process mechanization for the tillage and soil preparation,planting and fertilizing,field management and harvesting are necessary conditions for the standardization of soybean origin traceability.In order to solve these problems,soybean and soil samples from Beian Administration Bureau of Heilongjiang Farms and Land Reclamation Administration,and Nenjiang Northern Agricultural Development Co.Ltd of COFCO were taken as the research objects,and the organic composition and mineral element contents of soybean and soil samples from the two areas were determined through 2015 and 2016.By using Variance analysis,Principal component analysis,Cluster analysis and Discriminant analysis,3 organic components and 7 mineral element contents were selected as the traceability characteristic indexes.An soybean origin discriminant model was built,and combined with Dimension specification for linear discriminant analysis,the Discriminant system of soybean origin was developed based on Support vector machine algorithm.Results and conclusions were listed as following:(1)Four organic components contents(protein,fat,ash and soluble sugar)from 97 soybean samples from the experimental fields of two main soybean production areas were determined through 2015 and 2016.Statistic methods including Variance analysis,Principal component analysis,Cluster analysis and Discriminant analysis were used to explore the characteristic indexes of organic components for soybean origin traceability.The results indicated that the Overall correct discriminant rate and Overall correct cross-validation rate for soybean origin reached 86.0% based of the organic component discriminant model.(2)113 soybeans and soils samples from the experimental fields of two main soybean production areas through 2015 and 2016 were selected as the research objects.52 minerals elements from soybean and soil were determined,and Statistic methods including Variance analysis,Principal component analysis,Cluster analysis and Discriminant analysis were used toexplore the characteristic indexes of minerals elements for soybean origin traceability.The results indicated the Overall correct discriminant rate for two soybean origin reached 93.2%based on the 8 mineral elements(Mg,Mn,Sr,La,Gd,Tb,Hf,Ti,and other mineral elements)discriminant model.(3)112 soybeans and soils samples from the experimental fields of two main soybean production areas through 2015 and 2016 were selected as the research objects,and 56 random samples in 2014 were verification objects.Seven mineral elements including Mn,As,Sr,La,Nd,Tb,Hf,and three organic component,including protein,fat,and soluble sugar were selected as the characteristic index,and Fisher discriminant model was built for soybean origin traceability.According to the Dimension specification of linear discriminant analysis and Support vector machine(SVM),the discriminant rate of soybean origin was 94.6%,which was better than92.9% of the linear discriminant model.(4)The model based on MVC system was built on EF web page framework,the database for mineral elements and organic components of soybean was established,and the discriminant system for soybean origin based on Support vector machine algorithm was developed.By using the system for soybean origin discrimination,the correct discrimination rate was 97.5%,which was higher than 90% of the linear discriminant model.The established of the discriminant system for soybean origin traceability has a relatively high discrimination rate,which lays a theoretical foundation for the future development of soybean origin traceability.更多还原