Estimation for soluble solid content in Hetian jujube using hyperspectral imaging with fused spectral and textural Features

Soluble solid content (SSC) is an essential indicator for guiding the harvest period and evaluating the taste of fruit. Traditional measurement, such as refractometer method, is destructive and needs a certain time for solution preparation. This study develops a nondestructive and rapid method for SSC quantification in jujube using hyperspectral imaging (HSI) with wavelength range from 1003.22 to 2512.97 nm. Firstly, utilizing a higher flexibility for constructing multiple loss functions to improve accuracy and reduce spectral redundancy simultaneously, the characteristic wavelength selection framework based on multiobjective evolutionary algorithm by decomposition (MOEA/D) is constructed. Secondly, to explore the spatial information, gray level co -occurrence matrix (GLCM), local binary pattern (LBP) and Gabor filter are used to extract texture features and enrich the feature information. Finally, the extreme gradient boosting (XGBoost) model based on ensemble learning is established according to acquired spatial -spectral features. In comparison to traditional wavelength selection methods using only spectral information, the characteristic wavelengths selected by MOEA/D present a superior performance with (R2P=0.8760, RMSEP=0.8114, RPD=2.8392), and the R2P score increases by 0.03-0.05 on XGBoost model. Moreover, after the involvement of spatial texture features, the prediction performance is improved further with (R2P=0.9061, RMSEP =0.7031, RPD=3.2630). Thus, combining characteristic wavelengths and texture features, the experimental results prove the effectiveness for SSC prediction in Hetian jujube. Overall, this study facilitates the application of HSI-related nondestructive quality assessment for fruit. Furthermore, it also provides valuable reference for producers and manufacturers about the harvest period guidance and storage suggestion of foods and other agricultural products.