Prediction of soluble solids content in Xing'an Chinese pear-leaf Crabapple using NIR spectroscopy and CNN

Xing'an Chinese Pear-Leaf Crabapple is a geographical indication product of Inner Mongolia. Soluble solids content (SSC) is one of the primary indicators for fruit quality. However, traditional laboratory chemical testing is destructive and time-consuming, so this study used near-infrared spectroscopy (NIRS) to predict SSC of Chinese Pear-Leaf Crabapple. To reduce the noise in the field spectra data, four preprocessing methods including Savitzky-Golay smoothing (SG), standard normal variate (SNV), first derivative (D1), and baseline correction (Baseline) were applied. And competitive adaptive reweighted sampling (CARS), successive projections algorithm (SPA), and uninformative variable elimination (UVE) were employed to select the key wavelengths related to SSC. Besides, the prediction accuracy of three modeling methods, support vector machine (SVM), random forest (RF), and convolutional neural network (CNN) were compared. The results indicated that the D1 + CARS+CNN model demonstrated the best predictive performance with an R² and RMSE of 0.911 and 0.137, respectively. Compared with D1 + CARS+SVM and D1 + CARS+RF models, the R² was improved by 6.7% and 4.6%, respectively. And the spectral dimension was reduced by 39.52%. These results provide methodological support for the quantitative analysis of SSC in the field.