Mapping moisture contents in grass carp (Ctenopharyngodon idella) slices under different freeze drying periods by Vis-NIR hyperspectral imaging

- Freeze drying was conducted on grass carp slices.
- Moisture content in dried fish products was predicted using hyperspectral imaging.
- Partial least squares regression was applied to develop quantitative models.
- Selected wavelengths were used to build regression models with weighted values.
- The model built using weight-based selected wavelengths was the optimised one.

The Vis-NIR hyperspectral imaging system (400-1000 nm) was used to determine moisture content in grass carp slices under diverse freeze drying periods. Partial least squares regression (PLSR) models were developed based on the original spectra (381 wavelengths) and that pre-treated by multiplicative scatter correction or standard normal variate and nine wavelengths (414, 490, 520, 563, 580, 593, 634, 709, and 972 nm) selected using regression coefficients from the PLSR models. The simplified PLSR model based on the selected wavelengths achieved good prediction results (Rc2 = 0.9149, Rcv2 = 0.8974, RP2 = 0.9021; RMSEC = 55.0 g/kg, RMSECV = 61.1 g/kg, RMSEP = 56.1 g/kg). In order to further improve the performance of the simplified model, weighted values (1.75, 1.33, 1.85, 2.65, 1.13, 2.80, 1.35, 1.00 and 1.00) obtained from the regression coefficients method were applied to the selected wavelengths, based on which, the optimised PLSR model was finally developed (Rc2 = 0.9416, Rcv2 = 0.9278, RP2 = 0.9117; RMSEC = 45.5 g/kg, RMSECV = 51.2 g/kg, RMSEP = 56.3 g/kg), indicating different importance of the selected wavelengths in model development. Finally, the optimised model was used to generate the distribution map of moisture content in the samples under various freeze drying periods.