基于改进LLE算法的山东烟叶香型风格分析研究

Analysis of aroma type of leaf tobacco in Shandong based on improved LLE algorithm

为实现烟叶风格的快速鉴别,根据烟叶近红外光谱高维、非线性、冗余的特点,在局部线性嵌入(Locally Linear Embedding,LLE)算法的基础上,提出了一种改进LLE非线性降维算法。采用具有确定香型风格特征的基准参比样品建立了香型风格投影模型和判别模型,并与PCA、LLE降维方法进行了比较。结果表明该方法能够有效的对烟叶香型风格进行快速鉴别,准确率较高。应用该模型对2013年山东6大生态产区200个烟叶样品进行了分析,分析结果与以往专家的感官评吸结果基本一致。

In order to quickly identify aroma type, an improved Locally Linear Embedding（LLE） dimension reduction algorithm was proposed in line with high-dimension, nonlinear, redundant features of near infrared spectra. A certain aroma type was rendered as benchmark sample A projection model and a discrimination model were established by rendering certain aroma type as control sample and PCA and LLE dimension reduction methods were used as comparison. Results showed that improved LLE effectively discriminated aroma type of tobacco with higher accuracy. The models were further proved effective when applied to identify 200 tobacco samples from six major tobacco-growing areas of Shandong province and showed well consistency with previous sensory evaluation results.