Study on Modeling the Effect of Three-Dimensional Fluorescence Spectrum of Predicting the Content of Peanut Oil Adulterated Soybean Oil
WEI Quan-zeng1, LIU Xue-ying1, WANG Zhi-jie1, DING Fang2
1. Food and Pharmacy College, Xuchang University, Xuchang 461000, China
2. Shangqiu Academy of Agricultural and Forestry Sciences, Shangqiu 476000, China
Abstract：To determine the content of adulterated peanut oil in soybean oil, the three-dimensional fluorescence spectrum data of soybean oil and peanut oil counterfeit were collected. Rayleigh scattering and Raman scattering were removed using the triangular internal interpolation method. Then the fluorescence spectra were processed using Savitzky-Golay. The Alternating trilinear decomposition (ATLD) and Parallel factor (PARAFAC) algorithms were used to predict peanut oil content. Meanwhile, after scattering and smoothing the three-dimensional fluorescence data of the different contents of counterfeit peanut oil. The emission spectrum corresponding to each excitation wavelength is decomposed by wavelet packet decomposition (WPD), and the wave packet coefficient of the lowest frequency band is used as the characteristic amount of fluorescence emission spectrum data. All the emission wavelengths were reconstructed according to the sequence number of excitation wavelengths, and the data were reconstructed into a first-order fluorescence spectrum data vector. Partial least squares (PLS) and artificial neural network (ANN) data models were constructed to predict the content of peanut oil in counterfeit products. The results indicated the regression coefficients R2 of PARAFAC, ATLD, WPD-PLS, and WPD-ANN were 0.898, 0.941, 0.961, and 0.981, respectively. Mean absolute deviation (MAD), mean squared error (MSE), and root mean squared error (RMSE) of the training set, verification set, test set, and all data of the WPD-ANN algorithm model were all small. The peanut oil content in counterfeit products was predicted using the WPD-ANN model. The percentage of samples with prediction deviation within ±5% was 82.5%. The peanut oil content prediction results by WPD-ANN, WPD-PLS, ATLD, and PARAFAC were compared and analyzed. The mean and median deviations of WPD-ANN and WPD-PLS models are near 0%, while the mean and median deviations of ATLD and PARAFAC models are far from 0%. Compared with the PARAFAC model, the ATLD model has faster convergence and smaller deviation. ATLD and PARAFAC models may be affected by nonlinear factors, and their prediction effect was inferior to that of WPD-ANN and WPD-PLS, while ANN and PLS were based on first-order data regression modeling after WPD and data reconstruction. ANN was a nonlinear model. Therefore, the WPD-ANN model has stronger prediction ability and smaller deviation for peanut oil content in counterfeit peanut oil. The WPD-ANN model was the best algorithm among the four algorithms for predicting peanut oil content in counterfeit peanut oil. This provides a research basis for quantitative analysis of adulterated edible oil.
魏泉增，刘雪影，王至洁，丁 芳. 基于三维荧光光谱预测大豆油掺假花生油含量的建模效果研究[J]. 光谱学与光谱分析, 2025, 45(07): 1906-1915.
WEI Quan-zeng, LIU Xue-ying, WANG Zhi-jie, DING Fang. Study on Modeling the Effect of Three-Dimensional Fluorescence Spectrum of Predicting the Content of Peanut Oil Adulterated Soybean Oil. SPECTROSCOPY AND SPECTRAL ANALYSIS, 2025, 45(07): 1906-1915.
