依据流形学习的局部线性嵌入对红松子品质近红外检测

Near-infrared Detection of Outer Pine Nuts by LLE Manifold Learning

根据红松子品质检测方面技术缺失的现状,提出了运用近红外技术建立松子光-化学模型的解决办法;在对比流形学习有效保留高维数据的低维特征的优势和近红外传统降维方法主成分分析对非线性结构不敏感问题后,提出了具有能够捕捉高维空间中低维流形功能的局部线性嵌入-高斯过程(LLE-GP)方法,用于解决传统线性主成分分析(PCA)方法可能损失有用信息的缺陷;使用变量标准化(SNV)与Savitzky-Golay平滑方法进行预处理后,使用局部线性嵌入-高斯过程方法对数据进行分类建模。运用近红外光谱仪采集的松子数据,对这一算法进行验证,结果表明:局部线性嵌入-高斯过程分类模型,可以良好的使用在品质检测分类建模中。

The solution to establish the pine-light-chemical model using near-infrared technology was proposed.The advantage of the low-dimensional feature of the manifold learning to effectively preserve high-dimensional data was compared with that of the near-infrared traditional PCA dimensionality reduction method.From the sensitive problem,the LLE-Gaussian process method with the ability to capture low-dimensional manifolds in high-dimensional space was proposed to solve the defect that the traditional linear PCA method may lose useful information.The SNV and Savitzky-Golay methods were used for preprocessing.The data are classified and modeled using the LLE-Gaussian process method,and the score is calculated to evaluate the LLE-Gaussian process method in modeling results.