本文利用拉曼光谱和化学计量学方法, 建立快速分类模型对大米进行区分。在使用最小二乘法对离散拉曼光谱进行多项式拟合去除荧光背景的前提下, 利用在第一次迭代过程去除大型拉曼峰和计算噪声电平的方法, 并且保留数据维数在原来的50%以下。获取精确的拉曼信号。再用主成分分析法(Principal component Analysis, PCA)对3种大米全波段的拉曼光谱进行降维分析, 线性判别方法(Linear discrimination analysis, LDA)对样品进行分类, 结果显示采用前两个主成分能达到93.8%的正确分类, 采用前三个主成分能达到97.9%的正确分类。优化之后的模型对于大米的判别分析具有很好的效果。

A quick identification and classification model of rice was built based on Raman spectra and Chemometric methods. Firstly, least square method was used to remove fluorescence background of discrete Raman spectroscopy. Secondly, we acquired precise Raman signal by large Raman peak remove during the first iterative process and noise level calculation. while keeping data dimension below 50%. Finally, Principal Component Analysis(PCA) was employed to reduce dimension analysis of the full-wave band of three kinds of rice and Linear Discrimination Analysis(LDA) was used to classify them. The result showed that accuracies of 93.8% and 97.9% were obtained by utilizing first two PCs and first three PCs separately. The optimized model had a good performance for rice classification.