基于不同波段的枸杞多糖及总糖高光谱成像检测

发布时间：2018-03-19 01:15

本文选题：高光谱图像　切入点：BP神经网络　出处：《食品科学》2017年08期 　论文类型：期刊论文

【摘要】：采用高光谱图像技术对枸杞多糖和总糖含量进行检测,并探寻其最适宜的光谱波段。首先采用多元散射校正、Savitzky-Golay平滑(S-G平滑)和标准正态变量变换3种常用光谱预处理方法对原始光谱进行预处理,并对结果进行对比,选择多元散射校正预处理方法,以消除散射的影响;然后分别基于相关系数的数值及不同范围波长的特性,选择有效波段、可见光波段、近红外波段及全波段图像的平均光谱反射值作为特征参量;最后建立基于不同特征参量的枸杞多糖和总糖含量的BP神经网络预测模型。结果表明:基于全波段条件下光谱信息所建立的预测模型最佳,枸杞多糖含量预测正确率为97.59%,相关系数为0.997 4,均方根误差为0.077 7,枸杞总糖含量预测正确率为100%,相关系数为0.996 8,均方根误差为0.250 6。因此高光谱无损检测枸杞多糖和总糖含量具有可行性。
[Abstract]:The content of Lycium barbarum polysaccharides and total sugar was detected by hyperspectral image technique. At first, three common spectral pretreatment methods, Savitzky-Golay smoothing S-G smoothing and standard normal variable transformation, are used to preprocess the original spectrum, and the results are compared. The multivariate scattering correction preprocessing method is chosen to eliminate the influence of scattering, and then based on the value of correlation coefficient and the characteristics of different wavelength ranges, the effective band and visible light band are selected, respectively. The average spectral reflectance of near-infrared and full-band images is taken as the characteristic parameter. Finally, BP neural network prediction model of Lycium barbarum polysaccharides and total sugar content based on different characteristic parameters is established. The results show that the prediction model based on spectral information is the best. The correct rate of predicting the content of Lycium barbarum polysaccharide is 97.599.The correlation coefficient is 0.997 4, the root mean square error is 0.077 7, the correct rate of prediction of total sugar content of Lycium barbarum is 100, the correlation coefficient is 0.996 8, and the root mean square error is 0.2506.Therefore, hyperspectral nondestructive detection of Lycium barbarum polysaccharides and Lycium barbarum polysaccharides. The total sugar content is feasible.
【作者单位】：河南科技大学食品与生物工程学院;
【基金】：河南省科技创新杰出青年资助项目(624420017) 河南省科技攻关项目(172102210256;172102310617)
【分类号】：R284.1;TP391.41

【相似文献】