FOOD SCIENCE ›› 2019, Vol. 40 ›› Issue (16): 274-280.doi: 10.7506/spkx1002-6630-20180619-355

• Safety Detection • Previous Articles     Next Articles

On-Line Detection of Mildew Degree of Maize Based on Spectral and Image Information Fusion

SHEN Fei, HUANG Yi, ZHOU Yuechun, LIU Qin, PEI Fei, LI Peng, FANG Yong, LIU Xingquan   

  1. 1. Jiangsu Collaborative Innovation Center for Modern Grain Circulation and Safety, College of Food Science and Engineering, Nanjing University of Finance and Economics, Nanjing 210023, China; 2. Nanjing Lingshan Grain Reserve Co. Ltd., Nanjing 211599, China; 3. College of Agriculture and Food Science, Zhejiang A&F University, Hangzhou 311300, China
  • Online:2019-08-25 Published:2019-08-26

Abstract: This work aimed to establish an on-line method for the detection of the mildew degree of maize based on data fusion of visible/near infrared (Vis/NIR) spectroscopy and machine vision technique. Irradiation-sterilized maize kernel samples were inoculated with spore suspensions of 5 different fungal strains, separately. Then the samples were stored in an artificial climate box (28 ℃ and 85% relative humidity) for 15 d until serious mildew occurred. After 0, 6, 9, 12 and 15 d, the characteristic spectral and image information of maize samples were collected online. Both spectral features and image color features were extracted and fused for the development pf qualitative and quantitative analysis models. The results showed that principal component analysis (PCA) could distinguish maize samples according to different mildew degrees. The overall recognition rate obtained by linear discriminant analysis (LDA) for the classification of samples with different mildew degrees was 91.1% based on spectral and image data infusion, which were 4.4% and 8.9% higher than single spectral and image information, respectively. The prediction of colony counts in maize samples using the partial least squares regression (PLSR) based on information fusion was also better and the coefficient of determination for the prediction set (Rp 2), root mean square error of prediction (RMSEP) and residual predictive deviation (RPD) value obtained were 0.894 1, 0.665 (lg(CFU/g)) and 3.06, respectively. These results indicated that spectral and image data fusion could improve the model precision and was feasible in on-line detection of fungal contamination levels in maize. In further studies, naturally infected maize samples and samples contaminated with more representative fungal strains should be considered to enhance the robustness and applicability of the calibration model.

Key words: maize, mold infection, mildew, visible/near infrared spectroscopy, image, data fusion, on-line detection

CLC Number: