Detection of Aflatoxin B1 in Maize Using Hyperspectral Imaging Combined with Dimensionality Reduction

This study proposes a novel method for discriminating whether aflatoxin B1 (AFB1) content in maize kernels exceeds regulatory limits. Hyperspectral images ($1100- 2000 \mathrm{~nm})$ of maize samples with varying AFB1 concentrations were acquired. After spectral pretreatment, dimensionality reduction was implemented separately via Principal Component Analysis (PCA) and Factor Analysis (FA), and data fusion of both algorithms was investigated to derive optimal inputs for modeling. Extreme Gradient Boosting (XGBoost) was used to established a discriminant model. Results demonstrate that under the PCA-FA fusion-based dimensionality reduction approach, the model achieved 98.33% prediction accuracy using six variables as inputs, indicating excellent discriminative performance. This study provides a new methodological approach for discriminating excessive AFB1 concentrations in maize kernels.