Evaluation of trunk borer infestation duration using MOS E-nose combined with different feature extraction methods and GS-SVM

Trunk borer cause serious damage to plants and it is hard to detect due to larvae mine inside. In this study, E-nose combined with gas chromatography and mass spectrometry (GC-MS) was employed to evaluate Semanotus bifasciatus infestation duration at 0 d, 30 d, 60 d and 90 d. GC MS result indicated that the most abundant components were 3-carene, alpha-pinene, beta-phellandrene, sabinene and longifolene. The correlation between E-nose sensor responses and VOCs was analyzed by ANNOVA-Partial least square regression (APLSR). Six different features derived from E-nose data were analyzed by principle components analysis (PCA) and grid search-support vector machine (GS-SVM) was used as an optimized classifier to discriminate pest infestation duration based on different feature dataset. The classification results based on wavelet entropy (WE) showed preferable classification performances in both calibration set (100%) and validation set (100%). GS-SVM was also applied to predict S. bifasciatus infestation duration. Result showed that the fitting correlation coefficients (R-2) value of calibration set and validation set were 0.9987 and 0.9980, while the root mean square error (RMSE) of which were 0.4506 and 0.4961, respectively. It could be concluded that E-nose is a potential technique for evaluating trunk borer infestation and pest management.