基于SVM的大豆油脂色泽近红外光谱分析  

Near-Infrared Spectroscopy Analysis of Soybean Oil Color Based on SVM

作　　者：王立琦[1] 崔月[1] 张欢[2] 刘飞[2] 李贵滨[1] 葛慧芳[1] 于殿宇[2] 

机构地区：[1]哈尔滨商业大学计算机与信息工程学院,哈尔滨150028 [2]东北农业大学食品学院,哈尔滨150030

出　　处：《中国粮油学报》2015年第8期115-119,共5页Journal of the Chinese Cereals and Oils Association

基　　金：国家自然科学基金(31271886);黑龙江省高校科技成果产业化前期研发培育项目(1253CGZH22);哈尔滨商业大学博士科研启动项目(12DL023)

摘　　要：针对罗维朋比色计在油脂色泽测定中存在的问题,提出了一种基于支持向量机(SVM)的大豆油脂色泽近红外光谱分析方法。首先采用C-SVM对3种不同罗维朋黄值的大豆油脂进行模式识别,设计出适合油脂色泽近红外光谱识别的SVM分类器,识别正确率达到100%。然后利用ε-SVR对不同罗维朋黄值的大豆油脂近红外光谱数据与罗维朋红值进行回归,分别建立了不同级别大豆油脂色泽的SVM校正模型,预测误差均在0.2个罗维朋单位以内。研究表明,利用近红外光谱技术实现油脂色泽的定性定量分析是可行的,为进一步实现油脂色泽在线监测和调控提供参考。The paper has been aimed at the problems of Lovibond tintometer in determination of oil color. The experiment has presented a near- infrared spectrum analysis method for soybean oil color detection based on SVM. First, three categories of soybean oil with different Lovibond yellow values were recognized by C - SVM. The SVM classifiers which might be suitable for near - infrared spectral recognition of soybean oil color were designed then. The recognition correct rate had achieved 100%. Second, for the oils with different Lovibond yellow values, the regres- sions between near - infrared spectral data and Lovibond red values were conducted by ε - SVR. The SVM correction models for different grade soybean oil color were established respectively. The prediction errors could be controlled within 0.2 Lovibond Unit. The research demonstrated that it would be feasible to use near - infrared spectrum tech- nology to realize qualitative and quantitative analysis of oil color. The method proposed in the paper can be a refer- ence for further implementing on - line monitoring and control of oil color.

关 键 词：近红外光谱 油脂色泽 支持向量机 

分 类 号：TQ646[化学工程—精细化工]