基于小波变换的苹果汁多光程近红外光谱信息提取研究

INFORMATION EXTRACTION OF MULTI-OPTICAL-PATH NIR SPECTRA FOR APPLE JUICE BASED ON WAVELET TRANSFORMATION

利用浸入式光纤采集鲜榨苹果汁分别在5mm、10mm、15mm和20mm光程下的透/反射近红外光谱,实现对苹果汁中糖度(可溶性固形物,SSC)和酸度(pH值)的定量预测.结果表明,SSC和pH具有不同的最佳光程长,分别为5mm和20mm.为了兼顾各待测量对象的浓度范围和各组分的最佳光程长,从而提高模型的性能,采用多光程光谱混合建模,研究了多光程光谱信息的提取方法.采用原始光谱直接展开所建的模型虽然能有效利用多光程光谱的信息,但增加了模型复杂度,致使建模时间增长.因此,提出了两种基于小波变换的信息提取方法,它们在高效提取多光程信息的同时,能显著缩短建模时间并简化模型.其中基于展开光谱的小波近似系数建立的模型性能最优,SSC和pH值模型的SECV值分别达到0.4761oBrix和0.0779.

The transmittance/reflectance NIR spectra of fresh apple juice were collected by immersed fiber at four optical path of 5mm,10mm,15mm,and 20mm. The NIR spectra were used to predict quantitatively the soluble solid contents （SSC） and pH value of apple juice. The results show that the optimal optical path lengths for SSC and pH are different,which are 5mm and 20mm,respectively. In order to consider the concentration ranges of different components and their optimal path lengths,multi-optical path spectra were used to construct the model to improve the model quality. The method for extracting information from multi-optical-path spectra was studied. The method based on directly unfolded spectra could effectively use the spectral information,but the model became more complex and the calculation time was very long. Two methods based on wavelet transformation were proposed to extract the information. They can extract the multi-optical path spectra information effectively,and at the same time the calculation time decreases significantly and the model is simplified. One of the methods that uses the approximate wavelet coefficient of the unfolded spectra to construct model has the best performance： SECV for SSC and pH value are 0.4761° Brix and 0.0779,respectively.