基于光谱融合的茶油真伪快速鉴别研究

Rapid identification of Camellia oil authenticity based on spectral fusion

中文关键词:  近红外光谱  拉曼光谱  光谱融合  茶油真伪鉴别

英文关键词:near infrared spectroscopy  Raman spectroscopy  spectral fusion  Camellia oil  authenticity identification

中文摘要:

      目的 建立基于光谱融合的定性分析模型, 实现高值茶油的真伪快速鉴别。方法 优化设备条件, 同时采集茶油的近红外光谱(near infrared spectroscopy, NIRS)和拉曼光谱(Raman spectroscopy, RS), 分别使用6种方法进行预处理, 优选4种方法来提取光谱特征波段, 并应用数据层和特征层策略融合多光谱信息, 通过比较验证不同模型的准确率和预测均方根误差(root mean square error of prediction, RMSEP)来评估效果。结果 单独使用NIRS经标准正态变换处理后的偏最小二乘判别分析结果最优, 准确率为0.8361, RMSEP为0.1060; 单独使用RS经二阶导数处理后的结果最优, 准确率为0.8443, RMSEP为0.1332; 经NIRS和RS融合后数据结果高于任意单一光谱结果, 其中数据层光谱融合模型准确率为0.8525, RMSEP为0.1270, 特征层融合后的模型效果较好, 最佳结果为基于核主成分分析下的支持向量机模型, 准确率达到0.9508。结论 光谱融合提升茶油掺伪定性鉴别准确率更高, 具有较好的应用前景。

英文摘要:

      Objective To establish the qualitative analysis model by spectral fusion, realize the rapid identification of high-value Camellia oil. Methods The equipment conditions were optimized and the near infrared spectroscopy (NIRS) and Raman spectroscopy (RS) of Camellia oil were collected simultaneously, 6 kinds of methods were used for pre-processing, 4 kinds of methods were preferred to extract the spectral feature bands, and data layer and feature layer were applied to fuse the multispectral information, and the accuracy and root mean square error of prediction (RMSEP) of different models were compared and validated to evaluate the results. Results The best partial least squares discriminant analysis results using NIRS alone after standard normal transform processing were 0.8361 for accuracy and 0.1060 for RMSEP; the best results using RS alone after second order derivative processing were 0.8443 for accuracy and 0.1332 for RMSEP; the data fused with NIRS and RS gave higher results than any single spectral results, with the data layer spectral fusion model achieving accuracy of 0.8525 and RMSEP of 0.1270, and the feature layer fusion model gave better results, with the best result being the support vector machine model based on kernel principal component analysis, with accuracy of 0.9508. Conclusion Spectral fusion improves the accuracy of qualitative identification of Camellia oil adulteration, and has a good application prospect.