HighlightsE-nose was employed for evaluation of Semanotus bifasciatus infestation based on four time-domain features.Plant VOCs were analyzed by GC-MS, and the results proved the feasibility of E-nose detection.PNN, BPNN, SVM, and PLSR were introduced to classify and predict Semanotus bifasciatus infestation numbers.Abstract. Trunk-boring insects such as Semanotus bifasciatus (Motschulsky) are difficult to detect because the larvae are hidden inside the trunks. In this study, the variation of volatile organic compounds (VOCs) in Platycladus orientalis after S. bifasciatus infestation was evaluated using an electronic nose (E-nose). VOCs from sample plants were observed with gas chromatography - mass spectrometry (GC-MS), and the results indicated that uninfected and infected groups differed both qualitatively and quantitatively, which proves the feasibility of E-nose evaluation. To extract features of the E-nose response signals, four feature extraction methods were applied, and their performances were compared based on linear discriminant analysis (LDA). Three classification models, including back-propagation neural network (BPNN), support vector machine (SVM), and probabilistic neural network (PNN), were established to identify the severity of infestation based on the optimal feature extraction method (75th second value). The classification results of BPNN, PNN, and SVM based on the calibration set were 96.43%, 91.07%, and 100%, respectively, and the results based on the validation set were 91.67%, 91.67%, and 100%, respectively. In addition, partial least squares regression (PLSR) and BPNN were used to predict the larvae density and achieved highly reliable results. It can be concluded that combining E-nose with GC-MS is a potential technique for evaluating trunk-borer infestation and can be used for pest management. Keywords: Electronic nose, Feature extraction, Pest evaluation, Semanotus bifasciatus, Volatile organic compounds.
Dispersion Model of Volatile Organic Compounds Based on RBF Neural Network
