Non-Destructive Three-Grade Assessment of Pepper Sunscald Using Hyperspectral Imaging and Spectral Feature Fusion Strategy

Pepper production in southern Xinjiang, China, is severely constrained by sunscald, a non-infectious physiological disorder caused by intense solar radiation and high temperatures that compromises fruit quality and yield stability. This study proposes a complementary multi-feature fusion framework integrating Principal Component Analysis (PCA), Two-Dimensional Correlation Spectroscopy (2DCOS), and Competitive Adaptive Reweighted Sampling (CARS) to enable accurate three-grade (healthy, slight, and severe) non-destructive classification of pepper sunscald using hyperspectral imaging. Hyperspectral images of “Xiangla 702” pepper fruits were acquired and preprocessed via wavelet transform denoising; three feature sets were extracted and classified using Support Vector Machine (SVM), Random Forest (RF), and K-Nearest Neighbors (KNN) under single-feature, pairwise fusion, and triple-fusion schemes. Ablation experiments demonstrated that triple-feature fusion consistently outperformed single-feature and pairwise strategies across all classifiers. The optimal PCA-2DCOS-CARS-SVM model achieved a prediction set overall accuracy of 97.2%, a macro F1-score of 0.963, and a Kappa coefficient of 0.956, with 100% recall for healthy and slight (early-stage) samples and 90.9% recall for severe samples. These findings confirm that complementary multi-feature spectral fusion enhances sunscald grading performance, offering a promising methodological reference for non-destructive assessment of physiological disorders in horticultural crops.