Abstract: In this study, hyper-spectral data of eggs were collected using a hyper-spectral imager, and their freshness, pH and viscosity were also measured with a vernier caliper, a pH meter and a viscometer, respectively. Characteristic wavelengths were selected using competitive adaptive reweighed sampling (CARS) and successive projections algorithm (SPA), respectively, for constructing a simple multiple linear regression model (MLR) between egg quality and hyper-spectral features. Moreover, another MLR model was also established based on secondary SPA extraction of the characteristic wavelengths selected by CARS. The performances of the three models developed were compared. The results showed that the validation set correlation coefficients of both the CARS model and SPA model were above 0.9, which were lower than that of the CARS-SPA model with a lower root mean square error (RMSE) and fewer characteristic wavelength bands. The correlation coefficients of the MLR model for egg freshness, pH and viscosity were 0.94, 0.95, and 0.95 respectively, and the RMSE were 6.36, 0.17, and 149, respectively. This study indicates that optimized MLP model could be obtained using CARS or SPA to extract characteristic wavelengths and be further optimized by their combined use for non-destructive prediction of egg quality with improved stability.

Key words: eggs, freshness, acidity, viscosity, hyper-spectral imaging, successive projection algorithm