Hollow ZIF-8-enhanced composite colorimetric sensor array for early visual and nondestructive detection of citrus fruits infested by Bactrocera dorsalis

Early identification of citrus infestation by Bactrocera dorsalis ( B. dorsalis ) is critical for postharvest quality control and field pest management. Early infested fruit have no obvious visual symptoms, and the concentrations of characteristic volatile organic compounds (VOCs) they emit are low and easily affected by humidity, which limits the sensitivity and field applicability of existing detection methods. A hollow ZIF-8 (H-ZIF-8) reinforced dye composite colorimetric sensor array (CSA) with machine learning was constructed for visual, non-destructive and high precision identification of infested citrus. H-ZIF-8, prepared via tannic acid (TA) assistance, was combined with nine chemically responsive dyes to form the dye/H-ZIF-8 array. Notably, hollow porous structure of H-ZIF-8 enhanced VOCs diffusion and enrichment, boosting array response. Compared with pure dye and dye/ZIF-8 arrays, dye/H-ZIF-8 arrays exhibited higher sensitivity, richer color fingerprints, and stronger humidity interference. Actual sample tests showed color response intensity and total Euclidean distance (ED) of infested citrus increased apparently with infestation time, reflecting VOCs fingerprint changes. Based on these VOCs fingerprint changes, principal component analysis (PCA) of GC-MS VOC profiles separated healthy from infested samples, explaining 66.30% and 19.86% of the variance. Subsequently, with Dung Beetle Optimizer (DBO) feature selection and Tree-structured Parzen Estimator (TPE) hyperparameter optimization, the Support Vector Classifier (SVC) model achieved 92.71% accuracy in classifying healthy and citrus samples infested for 1–6 days. This study proposes a MOF reinforced colorimetric sensing strategy for non-destructive, visual, and batch-mode citrus pest monitoring.