摘要
科克铁热克葡萄作为地理标志产品,其品质特征与产地环境密切相关,为实现对其快速无损的品质评价与地标保护,该研究基于高光谱成像技术,构建了科克铁热克葡萄品质预测与地标产品识别多任务分析方法。研究获取了新疆科克铁热克地标产区及周边非地标产区葡萄样本的400~1000 nm高光谱数据,通过滤波平滑、一阶导数、二阶导数、基线校准、标准正态变换、多元散射校正等方法进行光谱预处理,以优化数据质量,基于偏最小二乘回归筛选最佳的预处理方法并构建出可溶性固形物(soluable solid content,SSC)和总酸(total acidity,TA)的品质预测模型。其中基线校准预处理效果最优,SSC品质预测模型测试集决定系数(R^(2))为0.8902,均方根误差(root mean square error,RMSE)为0.6898%,TA品质预测模型测试集R^(2)=0.7689,RMSE=0.1050%。针对地标产品识别任务,采用偏最小二乘判别分析构建地标产品识别模型,其滤波平滑预处理的模型性能最佳,R^(2)_(X)为0.987,R^(2)_(Y)为0.825,模型预测指数(Q^(2))为0.810,最终实现95%的判别准确率。该研究将化学计量学模型结合高光谱成像技术同时实现了科克铁热克葡萄的品质预测和地理标志识别,为农产品地理标志保护和水果品质无损检测提供了一种积极的技术方案。
As a geographical indication product,the quality characteristics of Koktiyerek grapes are closely related to their production environment.To achieve rapid,non-destructive quality evaluation and geographical indication protection,this study utilised hyperspectral imaging technology to develop a multi-task analysis method for predicting the quality of Koketireke grapes and identifying geographical indication products.The study collected hyperspectral data in the 400-1000 nm from grape samples in the Koketireke geographical indication production area and surrounding non-geographical indication production areas in Xinjiang.Spectral preprocessing was performed using methods such as filtering and smoothing,first-order derivatives,second-order derivatives,baseline calibration,standard normalisation,and multi-spectral scattering correction to optimise data quality.Partial least squares regression was used to select the optimal preprocessing method and construct quality prediction models for soluble solids content(SSC)and total acidity(TA).Among these,baseline correction(BSL)preprocessing yielded the best results.The test set R-square(R^(2))for the SSC quality prediction model was 0.8902,with a root mean square error(RMSE)of 0.6898%,and the test set R^(2)for the TA quality prediction model was 0.7689,with a RMSE of 0.1050%.For the task of identifying geographical indication products,partial least squares discriminant analysis was used to construct a geographical indication product identification model.The model with Savitzky-Golay smoothing(SG)pre-processing performed best,with R^(2)_(X)=0.987,R^(2)_(Y)=0.825,and a model prediction index(Q^(2))of 0.810,ultimately achieving a discrimination accuracy rate of 95%.This study combined chemometric models with hyperspectral imaging technology to simultaneously achieve quality prediction and geographical indication identification for Koktiyerk grapes,providing an active technical solution for the protection of agricultural product geographical indications and non-destructive quality detection of fruits.
作者
崔希炜
黄恺文
江涛
刘志刚
刘岩
刘海
于航
CUI Xiwei;HUANG Kaiwen;JIANG Tao;LIU Zhigang;LIU Yan;LIU Hai;YU Hang(Am Incorporation for Metrology and Testing Technology Services,Suzhou 215000,China;School of Food Science and Technology,Jiangnan University,Wuxi 214122,China)
出处
《食品与发酵工业》
北大核心
2026年第4期261-270,共10页
Food and Fermentation Industries
基金
新疆典型地理标志农产品特征品质挖掘技术研究(天科研[2023]719号)。
关键词
高光谱成像
地理标志产品
品质预测
光谱预处理
科克铁热克葡萄
hyperspectral imaging
geographical indication products
quality prediction
spectral pretreatment
Koktiyerek grapes
