Application of hyperspectral imaging in the detection of aflatoxin B1 on corn seed
2022
Zhang, Jun | Xu, Binbo | Wang, Zhiying | Cheng, Fang
Contamination of aflatoxins has important effect on the quality of corn seed. In this study, hyperspectral imaging technology was used to detect aflatoxin B1 (AFB1) with different concentrations (100, 50, 30, 20 and 10 μg/kg) on corn seeds. The near-infrared hyperspectral data, which were collected from the corn seed samples, were processed by image segmentation method. Then, the spectra in the range of 460–929 nm in the region of seed embryo were extracted from the seed-embryo hyperspectral image. Next, the modeling performance of two schemes under four preprocessing methods (none, multiplicative scatter correction (MSC), standard normal variate (SNV), and 5–3 smoothing), four feature wavelength extraction methods (full wavelength, principal component analysis (PCA), X-loading, and successive projection algorithm (SPA)) and three chemometrics methods (k-nearest neighbor (KNN), linear discriminant analysis (LDA), and support vector machine (SVM)) were compared. The optimal result of the testing set is in the case of “5–3 smoothing” pretreatment, SPA method, and SVM model, with 84.1%, 77.8% and 87.3%, 83.0% accuracies of the training set and testing set respectively in scheme 1 and scheme 2. The optimal accuracy of the testing set obtained by “SPA method” is 3.5% higher than in the case of “full wavelength” method, reducing the model calculations and improving the model accuracy to a certain extent. These 10 wavelengths (460.41, 491.50, 594.72, 617.60, 679.77, 726.69, 739.00, 772.43, 926.10, and 929.03 nm) obtained by “SPA method” provide the possibility for real-time online monitoring AFB1 contamination on corn seeds in the future.
显示更多 [+] 显示较少 [-]