Aflatoxin is a highly toxic and carcinogenic substance with fluorescence characteristic. To explore the feasibility of detection the degree of aflatoxin contamination using hyperspectral imaging technology, we proposed a machine learning detection method based on support vector machine (SVM) combining band index and narrow band. First, five concentrations of aflatoxin solutions (10ug/L, 20ug/L, 50ug/L, 100ug/L and 10 mg/L) were prepared and dripped onto the surface of different peanut kernels. Next, hyperspectral images with 33 bands (400-720 nm) were acquired for each sample using a hyperspectral imaging system under 365 nm ultraviolet (UV) light. Then four fluorescence indexes including Radiation Index (RI), Difference Radiation Indexes (DRI), Ratio Radiation Index (RRI) and Normalized Difference Radiation Index (NDRI) were proposed. Finally, Fisher method was used to optimize and obtain a narrowband spectrum, and RBF-SVM model was used to recognize aflatoxin and make regression analysis on the degree of aflatoxin contamination. Experimental results showed that DRI index had the optimal performance, and the accuracy rate of 5-fold cross validation of SVM were 95.5% and the mean square error (MSE) and correlation coefficient R were respectively 0.0223 and 0.9785. Results of this paper are of positive significance for the online aflatoxin detection and grading of agricultural products.
Keywords: Aflatoxin; Band index; Fisher; Hyperspectral imaging; Narrow band; Support vector machine.
Copyright © 2020 Elsevier Ltd. All rights reserved.