EARLY DETECTION OF TRUNK BORER DAMAGE IN PLATYCLADUS ORIENTALIS PLANTS USING E-NOSE AND GC-MS

Zhenhe Wang,Jun Wang
DOI: https://doi.org/10.13031/aim.201800374
2018-01-01
Abstract:Early detection of forest pests infection may allow timely control and prevent further spread of pest damage. In this study, an electronic nose (E-nose) was employed to evaluate Semanotus bifasciatus damage severity in Platycladus orientalis standing plants in plantation forest in early period. Gas chromatography - mass spectrometer (GC-MS) was used to analyze volatile organic components‘ (VOCs) profile, the results indicated that both the components and amounts varied with different damage severities and this explained the feasibility of E-nose detection. The infection number of pests was chosen as damage severity evaluation index. In order to extract features of E-nose response signals, four feature extraction methods were introduced and compared by linear discrimination analysis (LDA). Back-propagation neural network (BPNN), support vector machine (SVM) and probabilistic neural network (PNN) were used for pest number classification based on the “75th second value”. The overall classification accuracies of BPNN, PNN and SVM were quite satisfying, the results of calibration set were 96.5%, 91.1% and 100%; of validation set, were 91.6%, 91.7% and 100%, respectively. Partial least squares regression (PLSR) and BPNN were applied to predict the number of S.bifasciatus and their results were compared by fitting correlation coefficients (R2) and root mean square error (RMSE), both the two regression models performed well. It could be concluded that E-nose is a potential technique for detecting S.bifasciatus early infection in P.orientalis plants.
What problem does this paper attempt to address?