FOOD SCIENCE ›› 2023, Vol. 44 ›› Issue (24): 316-322.doi: 10.7506/spkx1002-6630-20230822-158

• Safety Detection • Previous Articles     Next Articles

Fast Quantification of Phosphorus in Crude Soybean Oil by Near-Infrared Spectroscopy

WANG Xue, ZHANG Hairong, WU Dandan, WANG Weining, WANG Liqi, LUO Shunian, YU Dianyu   

  1. (1. Harbin University of Commerce, Harbin 150028, China; 2. Jiusan Food Co. Ltd., Harbin 150060, China; 3. College of Food Science, Northeast Agricultural University, Harbin 150030, China)
  • Online:2023-12-25 Published:2024-01-02

Abstract: The existing methods for the determination of phosphorus content are unable to regulate the addition of acid and base in the refining process of crude soybean oil through real-time monitoring. Therefore, a novel rapid method for determining the phosphorus content of crude soybean oil based on near-infrared spectroscopy was proposed in this study. It was found that standard normal variate transformation was more effective than two other spectral preprocessing methods evaluated for denoising the spectral data indicative of the phosphorus content in soybean crude oil. The characteristic absorption band of phosphorus was optimized by synergy interval partial least squares (SiPLS). A back propagation (BP) neural network prediction model of the phosphorus content in crude soybean oil was established with learning efficiency of 0.005 and 108 training cycles. The determination coefficient (R2), root mean square error (RMSE) and relative standard deviation (RSD) for the correction set were 0.979 7, 0.859 3 and 1.89%, respectively. The R2, RMSE and RSD for the validation set were 0.978 5, 0.963 8 and 2.15%, respectively. The above results showed that NIR spectroscopy can achieve rapid, accurate and non-destructive detection of the phosphorus content in, and provide a feasible method for the refining of crude soybean oil.

Key words: near-infrared spectroscopy; crude soybean oil; phosphorus content; synergy interval partial least squares; back propagation neural network prediction

CLC Number: