库尔勒香梨含糖量的近红外光谱检测模型研究

Study on Near Infrared Spectroscopy Detection Model of Sugar Content in Korla Fragrant Pear

以成熟期库尔勒香梨为研究对象,以香梨含糖量作为检测指标,使用近红外光谱仪采集香梨样本光谱数据,通过一阶差分、二阶差分、标准正态变量变换(SNV)、多元散射校正(MSC)等方法对原始光谱进行预处理分析,研究香梨糖分的近红外光谱响应,结果表明MSC方法更适合于香梨近红外光谱数据的预处理。使用SPXY算法将近红外光谱建模样本集按4∶1进行划分,并使用相关系数法提取12个特征波长变量。通过线性回归、偏最小二乘法(PLS)和支持向量机(SVM)等方法分别建立香梨含糖量的检测模型,并进行比较,PLS模型均方根误差(RMSE)为0.5457,预测精度(Precision)为0.9918,相关系数为0.5802,均优于其它两种预测模型。MSC+PLS预处理方法可用于库尔勒香梨含糖量快速、无损检测。

Taking the mature Korla fragrant pear as the research object and the sugar content of fragrant pear as the detection index,the spectral data of fragrant pear samples were collected by portable near-infrared spectrometer.The original spectra were preprocessed and analyzed by first-order difference,second-order difference,standard normal variable transformation(SNV),mul⁃tivariate scattering correction(MSC)and other preprocessing methods to study the near-infrared spectral response of fragrant pear sugar,the results show that MSC method is more suitable for the preprocessing of near-infrared spectral data of fragrant pear.The sample set of near infrared spectrum modeling is divided by 4:1 using SPXY algorithm,and 12 characteristic wavelength variables are extracted by correlation coefficient method.The detection models of sugar content in fragrant pear were established by linear re⁃gression,partial least squares(PLS)and support vector machine(SVM).The root mean square error(RMSE)of PLS model was 0.5457,the precision was 0.9918,and the correlation coefficient was 0.5802.MSC+PLS pretreatment method can be used for rapid and nondestructive detection of sugar content in Korla Fragrant Pear.