%0 Journal Article %T 基于高光谱和CARS-IRIV算法的‘库尔勒香梨’可溶性固形物含量检测 %A 梁琨 %A 刘全祥 %A 潘磊庆 %A 沈明霞 %J 南京农业大学学报 %D 2018 %R 10.7685/jnau.201709030 %X [目的]利用高光谱技术实现‘库尔勒香梨’可溶性固形物含量的有效无损检测具有重要意义,但是高光谱数据通常噪声明显,大量无关信息变量和冗余信息变量的存在降低了模型的预测精度。本文旨在探究对高光谱数据特征变量筛选的有效方法来实现‘库尔勒香梨’可溶性固形物含量的快速检测。[方法]以‘库尔勒香梨’可溶性固形物含量(SSC)为研究指标,利用高光谱成像技术采集样本400~1 000 nm波长的漫反射光谱,对样本感兴趣区域(ROI)的光谱进行预处理,分别采用竞争性自适应重加权算法(CARS)、迭代保留信息变量算法(IRIV)以及CARS-IRIV算法筛选特征变量,基于不同筛选方法分别建立偏最小二乘(PLS)与最小二乘支持向量机(LS-SVM)预测模型,以预测集相关系数(Rp)、预测均方根误差(RMSEP)和预测相对分析误差(RPD)值对模型进行评价。[结果]CARS-IRIV算法可以有效减少CARS算法提取的变量个数,并稳定模型预测精度。LS-SVM模型预测结果优于PLS模型,在LS-SVM模型中CARS-IRIV-LS-SVM预测精度最高,Rp、RMSEP和RPD值分别为0.889、0.300和2.823。[结论]CARS-IRIV是一种有效的高光谱特征变量筛选算法,在提高预测精度的同时简化了模型的运算,CARS-IRIV-LS-SVM模型结合高光谱成像技术可以对‘库尔勒香梨’SSC进行快速有效的无损检测。</br>[Objectives]It is very important to realize soluble solids content in ‘Korla fragrant pear’ with hyperspectral imaging technology,but hyperspectral imaging data usually include a lot of noises,and a large number of irrelevant information variables and redundant information variables existing in hyperspectral imaging data would reduce the prediction accuracy of models. In this paper,an effective method for selecting the characteristic variables of hyperspectral data was explored for rapid detection soluble solids content(SSC)in ‘Korla fragrant pear’.[Methods]The diffuse reflectance spectra of 400-1 000 nm wavelengths were collected by hyperspectral imaging,the spectra of the region of interest(ROI)was preprocessed,and then the competency adaptive reweighing sampling(CARS),the iterated retaining informative variables(IRIV)and CARS-IRIV were used to select the characteristic variables. Finally,partial least squares(PLS)and least squares support vector machines(LS-SVM)were proposed to develop models,respec-tively. The correlation coefficient(Rp),the root mean square error of prediction(RMSEP)and the predicted relative error(RPD)were used to evaluate the models.[Results]The results showed that the CARS-IRIV algorithm could effectively reduce the number of variables in the CARS algorithm and stabilize the accuracy of the model. The LS-SVM model predicted better than the PLS model,in the LS-SVM model,the prediction accuracy of the CARS-IRIV-LS-SVM model was the best(Rp,RMSEP and RPD were 0.889,0.300 and 2.823,respectively).[Conclusions]The study showed that CARS-IRIV was an effective method to filter the characteristic variables,which simplified the operation of the model while improving the prediction precision. It is a rapid,non-destructive and accurate detection of ‘Korla fragrant pear’ SSC based on hyperspectral imaging technology using CARS-IRIV-LS-SVM model %K 高光谱成像技术 %K 库尔勒香梨 %K 可溶性固形物 %K 竞争性自适应重加权算法 %K 迭代保留信息变量算法< %K /br> %K hyperspectral imaging technology %K Korla fragrant pear %K soluble solids content(SSC) %K competency adaptive reweighting sampling(CARS) %K iterated retaining informative variables(IRIV) %U http://nauxb.njau.edu.cn/oa/darticle.aspx?type=view&id=201804025