Impact of maturity and variety on the hyperspectral detection model for determining soluble solid content in fresh apricots

The hyperspectral technique is a non-destructive method for quantifying the organic matter content, and has emerged as a prevalent approach for measuring soluble solid content (SSC) in fruits. However, existing hyperspectral models often face challenges in accurately predicting new samples of different maturities and varieties, thus limiting their wide application. The aim of this study was to investigate the effects of maturity and variety on the hyperspectral detection model for SSC of fresh apricot, and to establish a robust global model. Standard normal variable transformation method was used to reduce the influence of uneven light distribution on spherical fruit. There are four data sets: 6-1-pre-ripe stage, 6-1-ripening stage, Wanghong-ripening stage, and Jinmei-ripening stage. Through the Mahalanobis distance-concentration gradient (MD-CG) method, representative samples (5, 10, 15) were selected from the fresh apricot data set of other maturity and varieties as representative and typical sample points, and included in the training set for model update. This greatly improves the stability of the model. By integrating representative samples from the other three datasets into the 6-1-ripening (6-1-R) training set, combined with feature wavelength selection and radial basis function neural network updates, the global model effectively predicted SSC (RPD > 1.4) for the remaining three categories of fresh apricot. The global model is mainly affected by the variation of fresh apricot varieties, and the influence of maturity on the model accuracy is relatively small. The results show that using MD-CG recalibration model as a rapid detection strategy can determine the content of SSC in fresh apricots of different maturity or varieties, thus providing a solid foundation for its industrial application.