基于高光谱技术的金银花硫含量快速检测模型建立（英文）

Development of a Predictive Model for Rapid Detection of Sulfur Content in Honeysuckle Based on Hyperspectral Imaging Technology

为实现金银花硫含量的快速无损检测,利用高光谱成像技术结合化学计量学方法,建立不同浓度硫磺熏蒸金银花快速检测模型。采用硫磺使用量为鲜质量的0%、0.5%、1%、1.5%四种硫熏梯度的金银花干燥样品,首先利用高光谱成像技术采集各组金银花光谱图像数据,并采用S_G(Savitzky-Golay)卷积平滑、多元散射校正(multiple scatter correct,MSC)和标准正态变量变换(standard normal variate transformation,SNV)3种方法对原始光谱进行预处理,得到S_G卷积平滑为最佳预处理方法。随后,对经S_G预处理后的光谱信息分别进行Fisher判别分析(Fisher discriminate analysis,FDA)与核Fisher建模分析(kernel Fisher discriminate analysis,KFDA),得到KFDA具有更好的判别正确率(98.2%)。最后,全光谱数据具有量大、冗余信息的问题,采用了相关系数法(regression coefficients,RC)、Wilks和RC-Wilks三种方法对预处理后的数据进行特征提取,最终建立了RC-KFDA、WilksKFDA、RC-Wilks-KFDA三种判别模型。结果表明,经S_G卷积平滑预处理后的光谱信息,3种方法的判别正确率均为100%,使用RC-Wilks相结合提取特征波长的方法建立KFDA模型能够实现较短的计算时间(0.69 s)和较好的类间分布。因此,所建立的S_G-RC-Wilks-KFDA模型可以实现金银花不同硫含量的快速、有效、无损检测。

For rapid and non-destructive detection of sulfur content in honeysuckle,the flowers of Lonicera japonica Thunb.,hyperspectral imaging technology combined with chemometrics was applied to develop a predictive model for detecting sulfur-fumigated honeysuckle with different sulfur concentrations.Hyperspectral images of non-fumigated and sulfur-fumigated honeysuckle samples with four concentration gradients of0%,0.5%,1%and1.5%on a fresh mass basis were collected and preprocessed by Savitzky-Golay smoothing filter(S_G filter),multiple scatter correct(MSC)or standard normal variate transformation(SNV).S_G filter was selected as the optimal pretreatment method.Subsequently,the processed spectral data were used to establish models using either fisher discriminant analysis(FDA)or kernel Fisher discriminant analysis(KFDA),and the results showed that KFDA had a better discrimination accuracy of98.2%.Considering that the full-range spectral data contain a great deal of redundancy,the characteristic wavelengths were extracted by three different methods,regression coefficients(RC),Wilks criterion and RC-Wilks.As a result,the discriminant models,RC-KFDA,Wilks-KFDA and RC-Wilks-KFDA were developed.A comparison was made between these models,and the RC-Wilks-KFDA model was found to be the best one with the highest discrimination accuracy of100%,good classification efficiency and short running time of0.69s.Therefore,the S_G-RC-Wilks-KFDA model could allow fast,effective and nondestructive detection of sulfur content in honeysuckle.