手持式LIBS结合随机森林用于鱼产品的快速鉴别分析  

Rapid Identification of Fish Products Using Handheld Laser Induced Breakdown Spectroscopy Combined With Random Forest

作　　者：闫文豪 杨晓莹[1] 耿鑫 王乐山 吕亮 田野[1] 李颖[1] 林洪[2] YAN Wen-hao;YANG Xiao-ying;GENG Xin;WANG Le-shan;Lü Liang;TIAN Ye;LI Ying;LIN Hong(College of Physics and Optoelectronic Engineering,Ocean University of China,Qingdao 266100,China;College of Food Science and Engineering,Ocean University of China,Qingdao 266003,China)

机构地区：[1]中国海洋大学信息科学与工程学部物理与光电工程学院,山东青岛266100 [2]中国海洋大学食品科学与工程学院,山东青岛266003

出　　处：《光谱学与光谱分析》2022年第12期3714-3718,共5页Spectroscopy and Spectral Analysis

基　　金：国家重点研发计划课题(2019YFD0901701)资助。

摘　　要：中国是水产品生产和消费大国。由于不同鱼产品的品质和价格差距悬殊,近缘鱼类外观质地相似等特点,鱼产品掺假和错贴标签的现象频发,直接损害了消费者的消费和健康权益,因此实现鱼产品品种品质的快速检测具有重要的现实意义。激光诱导击穿光谱(LIBS)技术采用脉冲激光烧蚀样品表面产生激光诱导等离子体,通过探测等离子体的发射光谱实现待测样品元素组分的定性和定量分析,具有无需(或少量)样品预处理、多元素同时检测,分析速度快的优势,在食品快速检测分析方面具有很大的应用潜力。将LIBS技术结合随机森林(RF)算法用于不同种类鱼产品快速鉴别分析。首先对6种鱼肉样品进行压片处理,采用手持式LIBS分析仪采集其光谱数据,可探测到清晰的C、Mg、CN、Ca、Na、H、K、O等元素组分的特征谱线。将原始光谱数据进行归一化预处理,采用主成分分析方法(PCA)进行聚类,发现海水鱼和淡水鱼样品可以区分,而不同海水鱼之间和不同淡水鱼之间的样品则难以有效区分,说明PCA方法对鱼肉LIBS光谱分类能力有限。之后采用非线性的随机森林算法建立分类模型,经过优化RF模型的决策树个数与决策深度,得到鱼肉样品的整体识别正确率为90%。为进一步提高模型识别精度和分析效率,通过RF模型输出的变量重要性进行光谱特征提取,识别正确率提高到94.44%,且模型输入变量由23431个减少到597个,模型运算时间显著降低。表明RF模型结合变量重要性提取可以很好地将LIBS光谱中变量重要性高、对分类贡献大的弱信号提取出来,有效剔除了谱线噪声、背景、以及其他不相关变量的干扰,提高模型的识别精度和分析效率。也验证了手持式LIBS设备结合机器学习方法用于市场鱼产品快速鉴别分析的可行性。China is a big country of aquatic products production and consumption.Due to the great quality and price gap between the fish products from closely related species,the phenomena of adulteration and mislabeling of fish products have occurred frequently,which greatly encroached on the consumers’legitimate rights.Therefore,it is important to realize a rapid detection of the variety and quality of fish products.Laser-induced breakdown spectroscopy(LIBS)utilizes a pulsed laser to ablate the sample surface and generate a laser-induced plasma.Then the emission spectrum from the plasma is used for a qualitative or quantitative analysis of the elemental components of the sample.LIBS has shown great potential to be used in the food fast detection field with no or minimal sample preparation,multi-elemental analysis,and rapid detection capabilities.This paper applied LIBS combined with the random forest(RF)method to rapidly identify different fish products.Firstly,six fish samples were prepared into pellets,and the LIBS spectra were acquired using a handheld LIBS device.Clear spectral lines of C,Mg,CN,Ca,Na,H,K and O can be observed in the hand held-LIBS spectrum.After normalization of the raw spectral data,the principal component analysis(PCA)was used for clustering,and it was shown that the salt water fishes and freshwater fishes could be distinguished.In contrast,the different types inside the saltwater fishes or freshwater fishes can hardly be distinguished,indicating a limited capability of PCA method for the classification.Then,a nonlinear RF method was used to build the classification model.After optimizing the model parameters,including the decision tree number and the maximum depth,the RF model got an overall classification accuracy of 90%.In order to further improve the classification accuracy and efficiency,a feature selection method was performed by utilizing the variable importance of the RF model.It was shown that after feature selection,the classification accuracy was improved to 94.44%,and the number of inp

关 键 词：鱼产品 快速鉴别 手持式LIBS 随机森林 特征提取 

分 类 号：O433.4[机械工程—光学工程]