分子光谱技术结合深度学习模型识别食用植物油种类

为实现对食用植物油的快速无损识别，采用衰减全反射-傅里叶变换红外光谱获取10种食用植物油样本的340份谱图数据，经过预处理消除光谱数据中的噪声与背景干扰，通过主成分分析降维特征提取3个主成分，在此基础上构建KNN模型与基于SSA算法优化的BP神经网络模型，对植物油种类进行识别并对识别效果进行比较。结果表明：KNN模型的识别准确率可达97.7%;基于SSA算法优化的BP神经网络分类效果最佳，识别准确率达100%,而传统BP神经网络模型识别准确率仅为87.6%。综上，建立的分子光谱技术结合深度学习模型识别食用植物油种类的新方法，实现了对食用植物油种类的准确识别。

Identifying types of edible vegetable oil by molecular spectroscopic technology combined with deep learning model

To achieve rapid and non-destructive identification of edible vegetable oil, attenuated total reflection-Fourier transform infrared spectroscopy was used to obtain 340 spectral data of 10 edible vegetable oil samples. After preprocessing, the noise and background interference in the spectral data were eliminated. Three principal components were extracted by principal component analysis, and base on which, the KNN model and the BP neural network model optimized based on the SSA algorithm were constructed for identification and their effects were compared. The results showed that the recognition rate of the KNN model could reach 97.7%. The BP neural network model optimized based on the SSA algorithm, with a recognition rate of 100%, had the best classification effect, while the recognition rate of traditional BP neural network model was only 87.6%. In summary, a new method for identifying edible vegetable oil types using molecular spectroscopy technology combined with deep learning models can realize the accurate identification of edible vegetable oil types.