Integrating Olfactory and Visual Features for Precise Classification of Four Postharvest Mango Ripening Stages

Mango is a climacteric fruit whose postharvest quality is strongly affected by ripening stage. This study developed a low‐cost olfactory–visual feature‐level fusion framework to classify four postharvest mango ripening stages: mature green (MG), color break (CB), mature (M), and overmature (OM). During image‐related model development, 575 mango image records were used, while the fusion experiment included 100 paired Xiaotai mango fruits (25 per stage). Volatile‐related responses were acquired using a self‐developed multichannel metal oxide semiconductor gas‐sensing platform, and peel images were collected using a fixed industrial‐camera system. Eight gas‐sensor features and six hue, saturation, and value (HSV) image features were extracted from each paired fruit and concatenated into a 14‐dimensional fused feature set. Principal component analysis (PCA) was subsequently applied before logistic regression, retaining three components that explained 91.008% of the total variance. PC1 was mainly associated with gas‐sensor responses, whereas PC2 and PC3 were related to hue–brightness and saturation‐based visual information, respectively. Under repeated stratified fivefold cross‐validation, the best gas‐sensor‐only and HSV‐image‐only models achieved accuracies of 0.947 ± 0.040 and 0.941 ± 0.054, respectively. The final PCA‐reduced olfactory–visual logistic regression model achieved an accuracy of 0.975 ± 0.034 and a macro‐F1 of 0.975 ± 0.035. The confusion matrix showed that 99 of 100 samples were correctly classified, with the only error occurring between adjacent CB and M stages. Olfactory–visual fusion therefore improved classification over either single modality, while PCA provided a compact and interpretable feature representation. The proposed framework provides a low‐cost laboratory‐scale approach for nondestructive mango ripening‐stage classification.