Assessment of substrate properties of barley bran fermented by lactic acid bacteria using RGB reconstruction...

Rapid detection of substrate characteristics is crucial for precise control of the solid fermentation process. In this study, a method that reconstructs hyperspectral images from RGB images for detecting the characteristics of fermented barley bran substrates was explored. Using Competitive Adaptive Reweighted Sampling (CARS) to select key wavelengths, combined with a Multi-Scale Spectral Transformation Model (MST++), rapid and high-precision detection of moisture content and pH value in fermented barley bran was achieved by reconstructing key hyperspectral wavelengths from simple RGB images. The experimental results demonstrate that the CARS model accurately extracts critical wavelengths, while the hyperspectral data reconstructed by the MST++ model exhibit high accuracy in measuring moisture content and pH values, with a Mean Relative Absolute Error (MRAE) of 0.63, a Root Mean Square Error (RMSE) of 0.10, and a Correlation Coefficient (CC) of 0.95. Using multi-LED chromatic difference analysis to maximize the illuminant-metameric spectrum enhanced data discriminability and reduced metameric interference, achieving a 33 % improvement in MRAE and a 20.5 % reduction in RMSE over baseline models. The reconstructed spectral data also exhibited excellent performance in predicting substrate characteristics during fermentation, with correlation coefficients for moisture content and pH value predictions at 0.9207 and 0.8937, and RMSEP at 0.0419 and 0.1175, respectively. This technique could be helpful to improve the solid fermentation efficiency of agricultural and food products, and enhance the quality control of fermentation process.