基于可见/近红外透射光谱技术的红提糖度和含水率无损检测  

Non-destructive testing of red globe grape sugar content and moisture content based on visible/near infrared spectroscopy transmission technology

作　　者：高升 王巧华[1,2] GAO Sheng;WANG Qiao-hua(College of Engineering,Huazhong Agricultural University,Wuhan 430070,China;Key Laboratory of Agricultural Equipment in Mid-Lower Yangtze River Ministry of Agriculture,Wuhan 430070,China)

机构地区：[1]华中农业大学工学院,湖北武汉430070 [2]农业部长江中下游农业装备重点实验室,湖北武汉430070

出　　处：《中国光学》2021年第3期566-577,共12页Chinese Optics

基　　金：国家自然科学基金资助项目(No.31871863);湖北省自然科学基金资助项目(No.2012FKB02910);湖北省研究与开发计划项目(No.2011BHB016)。

摘　　要：本文研究基于可见/近红外透射光谱技术的红提糖度和含水率的无损检测方法。采集360个红提样本,并分别利用标准正态变量变换(Standard Normal Variable transformation,SNV)、SavitZky-Golay卷积平滑处理法(SavitZky-Golay,S_G)等光谱预处理方法处理后的数据建立PLSR模型,分别采用一次降维(GA、SPA、CARS、UVE)和二次降维组合(CARS-SPA、UVE-SPA、GA-SPA)7种数据降维方法对光谱进行特征变量提取,分别建立红提糖度和含水率的偏最小二乘回归算法(Partial Least Squares Regression,PLSR)和最小二乘支持向量机(Least Squares Support Vector Machine,LSSVM)含量检测模型并对比分析模型的优劣。结果表明:红提糖度和含水率的最优PLSR模型波长提取方法为GA-SPAPLSR,最优模型的预测集相关系数分别为0.958、0.938;红提糖度和含水率的最优LSSVM模型波长提取方法分别为CARS-SPA-LSSVM、UVE-SPA-LSSVM,最优模型的预测集相关系数分别为0.969、0.942;LSSVM所建模型的效果好于PLSR所建模型,但模型的运算时间较长。研究结果表明:基于可见/近红外技术无损检测红提糖度和含水率的方法可行,两种最优检测模型的预测精度均较高,都能满足检测要求。在不同应用下,可酌情选择不同模型,PLSR所建最优模型的运算时间较短,适合在线快速检测;LSSVM的检测性能最佳,可更加准确地检测红提糖度和含水率。In this paper,a non-destructive detection method for the sugar and moisture content of red globe grapes based on visible/near-infrared spectroscopy transmission technology is studied.The PLSR model is established by collecting 360 red globe grape samples by using spectral data processed by spectral preprocessing methods such as Standard Normal Variable transformation(SNV),SavitZky-Golay(S_G)and other spectral preprocessing methods respectively to determine the best spectral preprocessing method.Seven data dimensionality reduction methods of primary dimensionality reduction(GA,SPA,CARS,UVE)and secondary dimensionality reduction combinations(CARS-SPA,UVE-SPA,GA-SPA)are used to identify characteristic variables of spectra.PLSR and LSSVM detection models of sugar content and moisture content of red globe grape are established respectively,and the advantages and disadvantages of each model are compared and analyzed.The results show that the optimal PLSR model wavelength extraction method for red globe grape sugar content and moisture content is GA-SPA-PLSR,and the correlation coefficients of the optimal model are 0.958 and 0.938,respectively.The optimal LSSVM model wavelength extraction methods for red globe grape sugar and moisture content are CARS-SPA-LSSVM and UVE-SPA-LSSVM,respectively.The correlation coefficients of the optimal model are 0.969 and 0.942,respectively.The model built using LSSVM is better than that built using PLSR,but its operation time is longer.The results also show that the nondestructive detection method of red globe grape sugar and moisture content based on visible/near-infrared technology is feasible,and the detection accuracy of both two optimal detection models is high,which can meet detection requirements.Different models can be selected for different applications.The optimal model built by PLSR has shorter computation time and is suitable for online rapid detection.LSSVM has the best detection performance and can accurately detect red globe grape sugar and moisture content.

关 键 词：红提 糖度 含水率 可见/近红外技术 无损检测 

分 类 号：O657[理学—分析化学]