RESUMO
As economic globalization intensifies, the recent increase in agricultural products and travelers from abroad has led to an increase in the probability of invasive alien species. A major pathway for invasive alien species is agricultural quarantine materials (AQMs) in travelers' baggage. Thus, it is meaningful to develop efficient methods for early detection and prompt action against AQMs. In this study, a method based on the combination of odor detection of AQMs using ion mobility spectroscopy (IMS) and convolutional neural network (CNN) analysis for the identification of AQM species in luggage was developed. Two different ways were investigated to feed the IMS data of AQMs into the CNN, either as one-dimensional data (1D) (as a spectrum) or as two-dimensional data (2D) (as an IMS topographic map). The performances of CNN models were also compared to those of the commonly used classification algorithms: partial least squares discriminant analysis (PLS-DA) and soft independent modeling of class analogy (SIMCA). By doing gradient-weighted class activation mapping (Grad-CAM), the essential IMS feature regions from the CNN models to predict different AQM species were also identified. The results of this research demonstrated that the application of the CNN to the IMS data of AQMs yielded superior classification performance compared to PLS-DA and SIMCA. Especially, the CNN-2D model which utilized the IMS topographic map as input achieved the best classification accuracy both on the calibration and validation sets. In addition, the Grad-CAM method had an ability to detect critical discriminating spectral regions for different types of AQM samples, and could provide explanation for the CNNs' decision-making. Despite the inherent limitations of the present analytical protocol, the results showed that the method of IMS in combination with a CNN has great potential to be a complement for sniffer dogs and X-ray imaging techniques to detect AQMs.
