Portable NIR Prediction of Soluble Solids Content in Cherry Tomato Using Region-Guided Wavelength Selection and Sparse Bayesian Learning

Soluble solids content (SSC) is an important indicator of sweetness-related quality, maturity, and postharvest grading in cherry tomato. Destructive physicochemical measurements remain accurate but are labor-intensive and unsuitable for rapid batch assessment. This study developed an RGSW-SBL framework for predicting SSC in Zheyingfen cherry tomato using portable near-infrared (NIR) spectra. In this framework, region-guided stable wavelength selection (RGSW) was used to select informative wavelengths, whereas sparse Bayesian learning (SBL) served as the quantitative regression model for SSC prediction. RGSW integrates adaptive candidate waveband estimation with robust competitive wavelength screening, thereby retaining continuous spectral regions while reducing redundant and unstable variables. The prediction performance of SBL was compared with that of partial least squares (PLS) regression under full-spectrum and different wavelength-selection conditions. Model performance was evaluated using R2 and RMSE, with the principal test-set results reported in terms of R2 and RMSE. Under the training–validation–test evaluation protocol, RGSW-SBL achieved the numerically best internal-test result among the reported SBL combinations, with a test-set R2 of 0.847 and an RMSE of 0.369 °Brix using 157 selected wavelengths. The selected and high-contribution wavelengths were concentrated in chemically meaningful NIR regions related to C-H and O-H overtone absorption, sugar responses, and water-related tissue information. These results support RGSW-SBL as an interpretable framework for controlled-condition SSC prediction in cherry tomato, although multi-season, multi-cultivar, and multi-instrument validation remains necessary before deployment.