Detection and quantification of mold spores in wheat using hyperspectral imaging and data fusion techniques

Timely detection of mold severity is crucial for postharvest wheat quality and safety. This study developed a hyperspectral imaging (HSI) system integrated with chemometrics for the rapid, non-destructive quantification of mold severity via spore count analysis. Spectral data were preprocessed, and key features were extracted using competitive adaptive reweighted sampling (CARS), successive projections algorithm (SPA), and uninformative variable elimination (UVE). These features were then fused with texture information derived from gray-level co-occurrence matrix (GLCM) analysis. Three predictive models were constructed and evaluated: least squares support vector machine (LSSVM), random forest (RF), and convolutional neural network (CNN). The 1ST-CARS-GLCM-CNN model, integrating fused spectral-textural features, delivered superior performance, achieving a test-set coefficient of determination (R²) of 0.9904 and root mean square error (RMSE) of 0.1025, and enabled spatial distribution mapping for visual assessment. The HSI-chemometric approach effectively quantifies mold severity, providing a robust tool to enhance grain storage and food safety.