Single kernel aflatoxin and fumonisin contamination distribution and spectral classification in commercial corn

Aflatoxin and fumonisin contamination distribution in corn is non-homogeneous. Therefore, bulk sample testing may not accurately represent the levels of contamination. Single kernel analysis could provide a solution to these problems lead remediation strategies such as sorting. Our study uses extensive single aflatoxin (AF) (FM) measurements (i) demonstrate skewness, calculate weighted sums toxin for sample, compare those values measurements, (ii) improve classification algorithm performance. Corn kernels with natural (n = 864, from 9 samples) were scanned individually twice reflectance between ultraviolet–visible–near infrared spectrum (304 nm–1086 nm), then ground measured using ELISA. was non-homogeneous 1.0% 7) ?20 ppb (range 0 - 4.2×105 ppb), 5.0% 45) ?2 ppm 7.0×102 ppm). A sum calculated compared measurements. Average difference mycotoxin (AF 0.0 log(ppb), FM log(ppm), – levels) no systematic bias two methods, though considerable range ?1.4 0.7 log(ppb) AF ?0.6 0.8 log(ppm) FM. Algorithms trained on 70% classify (?20ppb) (?2ppm), while remaining 30% used testing. For aflatoxin, best performing stochastic gradient boosting model an accuracy 0.83 (Sensitivity (Sn) 0.75, Specificity (Sp) 0.83), both training set. fumonisin, penalized discriminant outperformed rest algorithms, 0.89 (Sn 0.87, Sp 0.88), 0.86 0.78, 0.87). The present improves foundations corn, can be applied high throughput screening. This demonstrates heterogeneous at level, comparing data traditional tests, utilizing UV–Vis–NIR spectroscopy system by level.