FOOD SCIENCE ›› 2018, Vol. 39 ›› Issue (16): 328-335.doi: 10.7506/spkx1002-6630-201816048

• Safety Detection • Previous Articles     Next Articles

Predictive Model for Detection of Maize Toxins with Sample Set Partitioning Based on Joint x-y Distance (SPXY) Algorithm and Successive Projections Algorithm (SPA) Based on Hyperspectral Imaging Technology

YU Huichun, LOU Nan, YIN Yong*, LIU Yunhong   

  1. (College of Food and Bioengineering, Henan University of Science and Technology, Luoyang 471023, China)
  • Online:2018-08-25 Published:2018-08-17

Abstract: In this paper, a rapid, non-destructive and accurate method for detecting the contents of aflatoxin B1 (AFB1) and zearalenone (ZEN) in maize using hyperspectral imaging was established by developing predictive models. Among 5 spectral pretreatments tested, standard normal variate (SNV) was found to be the best method for preprocessing the original spectral data. Sample set partitioning based on joint x-y distance (SPXY) algorithm combined with partial least squares regression (PLSR) was used to screen the differences in the predicted contents of AFB1 and ZEN from different calibration set samples, and 130 and 140 calibration set samples were selected for AFB1 and ZEN, respectively. On the basis of dimensionality reduction by the uniform spectral spacing (USS) method, the two variable extraction methods: successive projections algorithm (SPA) and competitive adaptive reweighted sampling algorithm (CARS) were compared. The results showed that 17 characteristic wavelengths were selected for AFB1 and ZEN, respectively and the PLSR model established based on fewer calibration set samples with the characteristic wavelengths had better predictive performance. The correlation coefficients (R2pre) and root mean square error of prediction (RMSEP) for AFB1 and ZEN content were 0.997 3 and 0.681 5, and 0.997 7 and 1.144 1, respectively, while the initial R2pre and RMSEP values were 0.994 4 and 0.984 6, and 0.991 6 and 2.320 9, respectively. The results indicated that despite reducing the reducing the complexity of the model, the predictive accuracy could be improved. Therefore, it is feasible to non-destructively predict the AFB1 and ZEN content in maize by applying hyperspectral imaging technology.

Key words: maize, hyperspectral imaging technology, non-destructive detection, aflatoxin b1, zearalenone

CLC Number: