拉曼光谱结合模式识别方法鉴别大米种类

为实现大米种类准确、快速的鉴别，选购72份大米样品，粉碎，采集粒度为100-140目米粉的拉曼光谱，对谱图数据进行去噪、归一化和特征提取后，综合运用主成分分析（PCA）、层次聚类分析（HCA）和支持向量机（SVM）三种方法对粳米、籼米和糯米进行聚类与模式识别研究。三种大米经PCA分析可直观地归为三簇，籼米和糯米可被区分开，但粳米与糯米、粳米与籼米不能区分。HCA结果表明粳米与籼米较难区分，糯米与其它两种米有较大差异，三种大米经HCA聚类分析准确率为81.94%。而采用SVM判别方法经10次运行后的平均识别率达98.86%。实验证明：拉曼光谱法结合支持向量机用于大米种类的分类与识别简单快速，在分析数据相对复杂的情况下，可快速建立分类模型并实现大米种类间的鉴定与识别。

Identification of Rice Varieties by Raman Spectroscopy and Pattern Recognition

In order to identify rice varieties accurately and quickly, 72 rice samples were purchased and Raman spectra of rice flour at 100-140 mesh were collected. After denoising, normalization and feature extraction, principal component analysis (PCA), hierarchical clustering analysis (HCA) and support vector machine (SVM) were used to cluster and pattern recognition of japonica, indica and glutinous rice. The three kinds of rice can be classified intuitively into three clusters by PCA analysis. Indica rice and glutinous rice can be distinguished, but japonica rice and glutinous rice, japonica rice and indica rice can not be distinguished. HCA results showed that japonica rice and indica rice were difficult to distinguish, glutinous rice was different from the other two kinds of rice. In addition, the classification accuracy of three kinds of rice was 81.94%. The average recognition rate of 10 runs of SVM discriminant method was 98.86%. Experiments showed that Raman spectroscopy combined with support vector machine was simple and fast for rice variety classification and recognition. Under the condition of relatively complex data analysis, classification model can be established quickly and identification and recognition among rice varieties can be realized.