【作者】 孙永； 刘楠； 王珊珊； 杨敏； 孙国辉； 曹荣； 周德庆；

【Author】 Sun Yong;Liu Nan;Wang Shanshan;Yang Min;Sun Guohui;Cao Rong;Zhou Deqing;Yellow Sea Fisheries Research Institute,Chinese Academy of Fishery Sciences;

【机构】 中国水产科学研究院黄海水产研究所；

【摘要】 仿刺参的地理原产地对其市场价值有着决定作用。食品溯源研究领域,红外光谱由于其采集速度快、成本低、对分析样品无损等独特优点得到广泛应用。数据融合技术是将多个传感器或者不同来源数据进行综合处理,充分利用信息的相关性和互补性实现更为全面准确的识别或者决策的新技术。为更准确鉴别仿刺参产地,本研究尝试了运用不同数据融合方法对获得的不同产地仿刺参近红外和中红外数据进行融合,以进一步提升仿刺参地理溯源机器学习模型的分类性能。研究结果表明,在数据水平上,近红外和中红外光谱数据的融合,由于变量剧增而实例不变导致了模型过拟合,相对于单独使用近红外或中红外数据的模型,分类性能出现下降;特征水平上,基于SHAP的选择特征不能完全涵盖光谱信息,造成模型性能下降;而通过PCA提取主成分特征后再融合,可以显著提升模型性能,使建立的lightGBM模型的分类准确率提升至97.44%。综上所述,基于PCA特征提取的红外光谱数据融合能够更准确地鉴别不同地理来源的仿刺参,可以为提升其他食品溯源准确率提供参考。更多还原

【Abstract】 The geographical origin of sea cucumber Apostichopus japonicus plays a decisive role in its market value.Infrared spectroscopy is widely used in food traceability research due to its unique advantages such as fast collection speed,low cost,and non-destructive analysis of samples.Data fusion technology is a new technology that comprehensively processes multiple sensors or data from different sources,and makes full use of the correlation and complementarity of information to achieve more comprehensive and accurate identification or decision-making.In order to more accurately identify the origin of sea cucumber,this study tried to use different data fusion methods to fuse the obtained near-infrared and mid-infrared data of sea cucumber,so as to further improve the classification performance of sea cucumber geographical traceability machine learning model.The results of the study show that at the data level,compared with models using near-infrared or mid-infrared data alone,the fusion of near-infrared and mid-infrared spectral data leads to model overfitting due to the dramatic increase in variables while the instances remain unchanged;At the feature level,the selected features based on SHAP cannot fully cover the spectral information,resulting in a decline in model performance;however,the extraction of principal component features through PCA and then fusion can significantly improve model performance.The classification accuracy of the established lightGBM model to 97.44%.In summary,the fusion of infrared spectral data based on PCA feature extraction can more accurately identify sea cucumbers from different geographical sources,and can provide a reference for improving the accuracy of other food traceability.更多还原