Data Fusion Strategy for Nondestructive Detection of Aflatoxin B1 Content in Single Maize Kernel Using Dual-Wavelength Laser-Induced Fluorescence Hyperspectral Imaging

Abstract

Aflatoxin B1 (AFB1) is the most widespread, toxic, and harmful mycotoxin, and maize is highly susceptible to AFB1 contamination, posing significant risks to human and animal health. Therefore, precise detection of AFB1 is essential to ensuring food safety. In this study, we used fluorescence probe technology to track the infection process of Aspergillus flavus in maize, confirming the uneven distribution of AFB1 and proposing the use of a “full-surface scanning” spectral information acquisition mode to improve detection accuracy. Therefore, we developed a full-surface fluorescence hyperspectral imaging system with high excitation/emission characteristics, combining dual-wavelength laser-induced fluorescence hyperspectral imaging and data fusion strategy to enable nondestructive detection of AFB1 in individual maize kernels. To address fluorescence crosstalk between maize substance and AFB1, we analyzed three-dimensional fluorescence spectra of healthy maize and pure AFB1 samples, identifying 360 nm and 405 nm as the optimal excitation wavelengths for AFB1 detection in maize. Furthermore, a prediction model for AFB1 content was constructed by combining different levels of data fusion strategies with a partial least squares (PLS) regression algorithm. The results showed that the dual-wavelength data fusion model was superior to the single-wavelength model. Specifically, the decision-level fusion model based on the characteristic wavelength selected by competitive adaptive reweighted sampling (CARS) achieved the best predictive performance (Rp = 0.83). This approach provides a new method for quantitative detection of AFB1 and lays the foundation for the advancement of AFB1 detection technology to enhance food safety.