Enhanced prediction of soluble solids content and vitamin C content in citrus using visible and near-infrared spectroscopy combined with one-dimensional convolutional neural network

The soluble solids content (SSC) and vitamin C (VC) content are two important quality indicators, which are of great significance for evaluating the quality, nutritional value, and market value of citrus. However, up to now, it hasn’t been achieved simultaneous nondestructive detection of these two internal quality attributes. Therefore, this study proposes a strategy for the nondestructive detection of SSC and VC content. This includes the use of visible and near-infrared (Vis-NIR) spectroscopy combined with partial least squares regression (PLSR) models built with different preprocessing methods and two effective wavelength selection methods, and compares the prediction accuracy with the one-dimensional convolutional neural network based on data augmentation (1D-CNN DA ) for detecting SSC and VC content in citrus ( C. reticulata 'Ai Yuan 38' ). The prediction results indicated that the 1D-CNN DA performed best, achieving correlation coefficient ( R p ) of 0.96, root mean square error ( RMSEP ) of 0.35 %, and relative percentage deviation ( RPD ) of 3.87 for SSC, and R p of 0.81, RMSEP of 2.81 mg/100 g, and RPD of 1.72 for VC. This research offers valuable insights into the use of Vis-NIR spectroscopy to enable simultaneous nondestructive detection for SSC and low concentration substances (such as VC) in other fruits.