Rapid detection of moisture content in solid-state fermentation by near-infrared spectroscopy combined with dbiPLS-SPA

Near-infrared spectroscopy (NIR) as an ideal tool was applied to measure moisture content in solid-state fermentation (SSF) of protein feed. To improve the detection precision and stability in determination of the moisture content in SSF by use of the NIR technique. Firstly, the raw spectra of all fermented samples obtained were preprocessed by use of the first derivative (1st Der). Secondly, the several efficient spectral subintervals were selected by use of dynamic backward interval partial least squares (dbiPLS). Thereafter, the feature combination variables were further extracted by successive projections algorithm (SPA) from the several spectral subintervals selected. Lastly, the partial least squares (PLS) model was developed by use of the feature combination variables selected for the measurement of moisture content of SSF of protein feed. In model calibration, the PLS factors were determined by a cross-validation, and The performance of the final model was evaluated according to the root mean square error of prediction (RMSEP) and correlation coefficient (Rp) in the validation set. The experimental results showed that the optimal model was obtained with 8 combined variables included, and these efficient variables corresponded to 7312.75 cm-1, 5850.97 cm-1, 5893.40 cm-1, 8527.68 cm-1, 5634.98 cm-1, 9538.20 cm-1, 9634.62 cm-1 and 9515.06 cm-1, respectively. The result of the RMSEP and Rp were 1.1795% (w/w) and 0.9430 in the validation set, respectively. Finally, the superior performance of the dbiPLS-SPA model was demonstrated by comparison with four other PLS models. The results indicate that NIR spectroscopy can be successfully used for measurement of moisture content in solid-state fermentation. Additionally, it is necessary to select characteristic wavelength variables of near-infrared spectra in model calibration. The dbiPLS-SPA is an effective method of combined variable selection. It can effectively reduce the complexity and improve generalization performance of the detection model when NIRS technique is used for on-line detection of the process parameters of SSF.