Abstract
Aflatoxin B (AFB) is a very strong carcinogen. Maize flour and other cereals tend to produce this toxin when stored under unsuitable conditions. Rapid and accurate detection and classification of AFB concentration is important to ensure food safety. In this study, a novel method for classifying AFB concentration in maize flour was developed. Three groups of maize flour samples with different AFB concentrations (10, 20, and 30 ppb) and one group of control samples were prepared. The visible and short wave near-infrared (Vis–SWNIR) region (430–1000 nm) and long wave near-infrared (LWNIR) region (1000–2400 nm) hyperspectral images of all samples were obtained, and the spectra of 430–2400 nm were obtained after spectral pretreatment and fusion. Then, two characteristic wavelength selection algorithms, namely, between-class to within-class variance ratio (BWVR) and weighted between-class to within-class variance ratio (WBWVR), were proposed. Both algorithms can effectively extract a small number of wavelengths with the largest difference information in the full wavelength, which is conducive to the establishment of classification model. Based on three different classification models, namely, support vector machine (SVM), k-nearest neighbors (KNN), and decision tree (DT), BWVR and WBWVR achieved good effect in less than 10 characteristic wavelengths, especially the WBWVR algorithm. Finally, through the cross-validation of the samples in the three sample plates, the average classification accuracy of AFB concentration of maize flour based on SVM model reached 96.18% under 10 characteristic wavelengths selected by WBWVR, thereby achieving good detection results. This study provides a new algorithm for the key wavelength selection of hyperspectral images, and also provides a new approach for AFB concentration classification of maize flour.
Published Version
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call