Discrimination of citrus infested with Bactrocera dorsalis (Hendel) using an olfactory detection system based on MIPs-QCM sensor array and Bayesian optimized classification algorithms

Abstract

The characteristic substances in citrus volatile organic compounds (VOCs) are associated with the infestation of the Bactrocera dorsalis (Hendel), which provides a noninvasive evaluation method to discriminate infested citrus. This paper developed an olfactory detection system based on a quartz crystal microbalance (QCM) sensor array and classification algorithms to identify the citrus infested with B. dorsalis. Six characteristic substances, including d-limonene, myrcene, α-pinene, decanal, linalool, and β-ocimene, which vary significantly after B. dorsalis infestation were selected as template molecules to prepare molecularly imprinted polymers (MIPs) and modify QCM respectively. The experimental results show that the prepared MIPs-QCM sensors had sensitivity in the range of 0.043–0.070 Hz/(mg/m3), and their stability and reproducibility were above 93.1%. Four sensors that contributed to the classification were screened by a stepwise discriminant analysis. Afterward, Bayesian optimization was employed to optimize the hyperparameters. The accuracy of the optimized support vector machine (SVM) reached 94.17%. The olfactory detection system developed in this study enables the discrimination of citrus infested with B. dorsalis, which may have potential applications in the field of post-harvest treatment of citrus.