基于PCA-SVM的仿刺参产地溯源在线系统研究

Study on Apostichopus Japonicus Origin Identification Online System Based on PCA-SVM

仿刺参是具有极高经济价值的水产资源,是海水养殖产业的重要组成部分,研发出一种灵活、稳定、高效的仿刺参产地溯源方法对于水产养殖产业具有极强的现实意义。仿刺参主要有三种养殖方式,分别是底播增殖、圈养养殖和筏式养殖。不同产地采用不同的养殖方式,仿刺参的营养价值、药用价值和经济价值都存在着明显差异。不同产地初级生产者的构成不同,作为初级生产者的不同藻类与浮游生物体内的脂肪酸特征也各不相同,通过食物链的传递,不同产地的仿刺参具有了不同的脂肪酸特征。气相色谱指纹图谱法是一种快速准确地食品产地溯源技术,碳稳定同位素质谱法不仅可以鉴别产地还可以区分出食品的营养价值。采集9个最具代表性产地的仿刺参样品,先利用Folch法对样品进行总脂提取,再通过气相色谱仪测定出各种脂肪酸的种类及其相对含量;最后使用稳定同位素质谱仪测定出每种脂肪酸各自的碳稳定同位素组成数据。使用单因素方差分析法对脂肪酸相对含量和脂肪酸碳稳定同位素组成数据进行显著性检验,各筛选出17种脂肪酸数据作为两个模型的输入。主成分分析(PCA)法可以降低数据的维度,聚合不同种脂肪酸数据的溯源特征,提高产地溯源的精度。支持向量机(SVM)是一种以结构化风险最小为目标的分类识别算法,具有优秀的泛化能力。研究结果表明,不同产地仿刺参的脂肪酸相对含量和脂肪酸碳稳定同位素组成数据存在明显差异。通过主成分变换后,脂肪酸数据的聚类特征更加明显,运用随机交叉验证法确定前6个主成分作为两个支持向量机分类器的输入。采用基于遗传交叉因子改进的粒子群优化算法(GPSO),以粒子不同K值各100次交叉验证的平均准确率作为其适应度,寻找支持向量机分类器模型的最优参数组合。最终计算得到脂肪酸相对含量产地溯源模型的最优参数组合为σ=6.247 599和C=14.313 042,平均准确率为79.49%;脂肪酸碳稳定同位素组成产地溯源模型的最优参数组合为σ=7.626 194和C=2.193 410,平均准确率为98.33%。对比交叉验证的结果,脂肪酸碳稳定同位素组成产地溯源模型具有更高的准确率和更强的泛化性能。在两个模型的识别结果不一致时,采用脂肪酸碳稳定同位素组成模型的识别结果。将实验室检测与互联网技术进行整合,构建了仿刺参产地溯源在线系统。实现了"互联网+产地溯源"的一体化溯源模式,为进一步开展食品产地溯源研究提供了科学依据和技术支撑。

Apostichopus japonicus,an important part of the mariculture,is a fishery resources with extremely high economic value.Therefore,it is of great practical significance for the mariculture to study a flexible,stable and efficient method to identify the origin information of apostichopus japonicas.There are three main aquaculture methods for the apostichopus japonicas,including bottom sowing culture,captive culture and raft culture.Different aquaculture methods are used in apostichopus japonicus of different producing areas,and there also exist great differences in the nutritional value,medicinal value and economical value from different producing areas.The compositions of primary producers vary from one provenance to another,and the characteristics of fatty acids in different algae and plankton as primary producers are also different.Through the transmission of the food chain,apostichopus japonicus from different producing areas have different fatty acid characteristics.Gas chromatographic fingerprint is a fast and accurate traceability technology for food origin.Carbon stable isotope ratio mass spectrometry can not only identify origin but also distinguish the nutritional value of food.Samples of apostichopus japonicus were collected from nine representative producing areas,and total lipid data were extracted by using the Folch method.Then determined the data of fatty acid kinds and relative content through gas chromatography.Finally,stable isotope ratio mass spectrometer was used to determine the data of fatty acids carbon stable isotope compositions.One-factor analysis of variance (ANOVA) was used to test the significance of the data of fatty acid relative content and fatty acids carbon stable isotope compositions,and then selected17 kinds of fatty acid data as inputs for the two models.The principal component analysis(PCA) method can reduce the dimensions of the data,and aggregate the origin characteristics of different fatty acids to improve the accuracy of origin identification model.Support vector machine (SVM) is a classification algorithm that aims to minimize structural risk and has good ability of generalization.The results indicated that there were significant differences in the fatty acids relative content and fatty acids carbon stable isotope compositions data of apostichopus japonicus from different producing areas.After the principal component analysis,the clustering characteristics of the fatty acids data were more obvious.With the cross-validation method,the first six principal components were determined as inputs of the two support vector machine classifiers.Applied the particle swarm optimization improved based on genetic crossover factor(GPSO),and the average accuracy of100 cross-validation results of different K values of the particle was used as the fitness to find optimal parameter combinations of the SVM classifier model.Finally,the optimal parameter combinations of fatty acids relative content model were σ= 6.247 599 and C =14.313 042,and average accuracy was 79.49%;the optimal parameter combinations of fatty acids carbon stable isotope compositions model were σ= 7.626194 and C =2.193 410,and average accuracy was 98.33%.Compared with the results of cross-validation,the fatty acids carbon stable isotope compositions origin identification model has higher accuracy and stronger ability of generalization.When origin identification results of two models were inconsistent,the results of fatty acids carbon stable isotope compositions model were used.The laboratory detection was integrated with Internet technology to build an apostichopus japonicus origin identification online system.The Integrated model of “Internet+origin identification” has achieved to provide a scientific basis and technical support for the further studies on origin identification of food.