A New Spectral Pretreatment Method for Detecting Soluble Solids Content of Pears Using Vis/ NIR Spectroscopy

Long Li,Yankun Peng,Yongyu Li,Cheng Yang,Xin Qiao

DOI: https://doi.org/10.13031/aim.202000495

2020-01-01

Abstract:Rapidly determining the soluble solids content (SSC) of fruit is significance for improving fruit value and farmer income. Vis-NIR spectra contain internal chemical information about the pear samples, which make it possible to evaluate the quality of fruit. However, when we use Vis-NIR spectroscopy to analyze complex heterogeneous mixture samples that exhibit sample-to-sample variability in physical properties, there are multiplicative light scattering effects, which are caused by the different optical path lengths due to the physical properties (sample size, particle size and shape, etc.) of different samples. It can “scale” the entire raw spectrum and lead to multiplication bias of the Lambert-Beer law. Further common linear multivariate correction methods, such as partial least-squares (PLS) and principal component regression (PCR), will also be invalidated due to the existence of multiplicative effect. Therefore, spectral preprocessing is required before the establishment of a multivariate linear correction model. In this paper, we utilized the spectral ratio combined with correlation analysis and stability competitive adaptive reweighted sampling (SR-CSCARS) method to eliminate the multiplicative scattering effect. The basic assumption of this method is that the multiplicative effect caused by physical change has a major effect on the spectral contribution of the mixture sample with the light scattering effect, and the sample was seen as the absorber in this paper. On the basis of this underlying assumption, multiplicative scattering correction was realized with two steps. First, the spectral ratio matrix was calculated by simple division operation, and then appropriate variable selection methods were used to obtain effective spectral ratio information, which showed improved correlation with the chemical concentration (SSC) in the pear samples. In the second step of the method, the correlation between the spectral ratio data and the chemical composition concentration was used to eliminate some of the irrelevant information, and then the SCARS algorithm was used to select the final effective information. Compared with other exiting methods (e.g., multivariate scatter correction (MSC), standard normal variate (SNV), and first derivative (FD)), SR-CSCARS provided optimal prediction result. In addition, this method does not depend on the ideal spectrum, so it can be extended to multispectral data processing.