非线性流行降维与近红外光谱分析技术的大米贮藏期快速判别

发布时间：2019-01-16 05:48

【摘要】：提出一种基于流行降维的近红外光谱技术快速判别大米贮藏期的新方法。采用近红外光谱仪获取陈年米和新米的反射光谱特征曲线,利用直接正交信号矫正法(direct orthogonal signal correction,DOSC)对原始光谱进行预处理,滤除光谱数据中与因变量Y矩阵无关的信号,以消除无关信息对后续特征变量建模精度的影响。采用Durbin-Watson和Run测试法定性分析光谱数据结构的非线性性,并利用增强偏残差图(augmented partial residual plot)定量分析大米光谱曲线的非线性程度。分别采用线性流行降维法包括主成分分析法(PCA)和多维尺度分析法(MDS)以及非线性流行降维法包括等距映射法(ISOMAP)、局部线性嵌入法(LLE)和拉普拉斯特征映射法(LE)提取预处理后光谱数据的本征变量,并结合核偏最小二乘方法(KPLS)建立本征变量与贮藏时间属性之间的耦合模型。实验用陈年米和新米的样本数均为200个,随机将训练集和测试集样本划分为300个和100个。通过比较各个模型的预测结果得出,基于ISOMAP非线性降维法提取的40个本征变量建立的回归模型预测效果最好,预测相关系数(R2P)、预测均方根误差(RMSEP)和预测相对分析误差值(RPD)分别为0.917,0.187和2.698。实验结果说明提出的方法对于大米贮藏期具有很好的鉴别能力,该研究为今后大米贮藏期的快速无损检测提供了科学的手段。
[Abstract]:A new method based on popular dimensionality reduction near infrared spectroscopy (NIR) was proposed to quickly identify the storage period of rice. The characteristic curves of reflectance spectra of aged and new meters were obtained by using near infrared spectrometer. The original spectrum was pretreated by direct orthogonal signal correction method (direct orthogonal signal correction,DOSC), and the signals independent of dependent variable Y matrix were filtered out in the spectral data. In order to eliminate the influence of independent information on the modeling accuracy of subsequent feature variables. The nonlinearity of spectral data structure was qualitatively analyzed by Durbin-Watson and Run, and the degree of nonlinearity of rice spectral curve was quantitatively analyzed by enhanced partial residuals (augmented partial residual plot). Linear epidemic dimensionality reduction method including principal component analysis (PCA) and multidimensional scale analysis (MDS) and nonlinear epidemic dimensionality reduction method (ISOMAP),) including isometric mapping (ISOMAP),) are used respectively. The local linear embedding method (LLE) and Laplace feature mapping (LE) are used to extract the intrinsic variables of the pre-processed spectral data, and the coupling model between the intrinsic variables and the storage time attributes is established with the kernel partial least squares method (KPLS). There are 200 samples of aged and new rice in the experiment. The training set and test set are randomly divided into 300 and 100 samples. By comparing the prediction results of each model, it is concluded that the regression model based on 40 intrinsic variables extracted by ISOMAP nonlinear dimensionality reduction method has the best prediction effect, and the prediction correlation coefficient (R2P) is the best. The root mean square error (RMSEP) and the predicted relative analysis error (RPD) were 0.917 ~ 0.187 and 2.698, respectively. The experimental results show that the proposed method has a good ability to identify rice storage period, and this study provides a scientific means for rapid nondestructive testing of rice storage period in the future.
【作者单位】：盐城工学院电气工程学院;四川农业大学机电学院;
【基金】：国家自然科学基金项目(31501221,31601227) 江苏省自然科学基金项目(BK20140467,BK20161310) 江苏省高校自然科学研究面上项目(13KJB210006) 盐城市农业科技指导性计划项目(YKN2014009,YKN2014010)资助
【分类号】：O657.33;TS210.7

【参考文献】