Study Of Pest Information For Tea Plant Based On Electronic Nose

Tea(Camellia sinensis)with great flavor and high content of beneficial substances,is the most widely consumed beverage aside from water.Especially in China,tea has been used as a daily beverage and crude medicine for thousands of years.Tea plant is grown in about 60 countries and is an important crop in many countries,such as China,Japan,Kenya and India.However,tea plants are easily attacked by pests,causing 5%?55%losses in tea production and 0.5 billion to 1 billion dollars in economic losses every year.In this study,Electronic nose(E-nose)and Gas Chromatography-mass Spectrometer(GC-MS)were employed.Tea plants attacked by different types of pests,the best detection time point for E-nose and tea plants with different attacked time,tea plant with different mass losses,tea plants attacked by two pests species were detected.The main conclusions are as follows:(1)Four feature extraction methods,feature extraction based on adsorption kinetic equation,polynomial function,exponential function and Gaussian function,were employed to deal with three E-nose dataset.The curve fitting figure and root-mean-square error(RMSE)were taken to evaluate the fitting performance of four functions.The results indicated that adsorption kinetic equation had the best fitting performance.Multi-layered perceptron(MLP)were applied for classification as well as chosing the best feature extraction method.The results indicated that the feature ectraction method based on adsorption kinetic equation was the best and good classification was obtained.(2)Tea plants attacked by different pest species were detected by E-nose and GC-MS.The results of GC-MS showed that volatile organic compounds(VOCs)emitted by tea plants attacked by different pest species was different,which indicated that it was possible for E-nose to detect them.Four dimensionality reduction methods(Principal Component Analysis(PCA),Locality Preserving Projections(LPP),Kernel Principal Component Analysis(KPCA)and Locally Linear Embedding(LLE))and three classification algorithms(Multilayer Perceptron Neural Network(MLPNN),Extreme Learning Machine(ELM)and Support Vector Machine(SVM))were employed and the best combination of dimensionality reduction method and classification algorithm was determined.The results showed that the combination of LPP and SVM was the best,and its correct discrimination rate was as high as 100%.(3)Tea plants with different time under attack were detected by E-nose and GC-MS.Furthermore,the performance of E-nose and GC-MS in different detection time points were also compared.In this study,we determined the time of tea plants under attack considering detection time point.The results of GC-MS showed that the VOCs emitted by tea plants influenced by both the detection time point and attacked time.The classification performances of E-nose based on various detection time points were compared and the best detection time point for E-nose detection was determined.Besides,ELM was employed for classification of tea plants with different times under attack and predicting the time of plant under attack based on the dataset obtained by E-nose detecting at the best detection time point.The results showed that the best detection time point of E-nose for tea plant was 12 noon,and ELM for classification and prediction all turned out good results,which indicated the feasibility of E-nose for detection of the time of tea plant under pest attack.(4)Tea plants with different damage severity were detected by E-nose and GC-MS,and the mass loss was taken as the evaluation index.The results of GC-MS indicated VOCs emitted by tea plants with different damage severities were different.The prediction performance of mass loss was compared with those of number of pest and time under attack based on Partial Least Squares Regression(PLSR)according to fitting correlation coefficients(R2)and RMSE respectively,and the results showed that the prediction performance of mass loss was better than the other two indexes.Then,three regression algorithms(PLSR,ELM and SVM)were applied to predict mass loss.The results indicated that these three algorithms all had good performances and SVM was the best one.It could be concluded that E-nose is a feasible technique for evaluating damage severity of tea plant and mass loss was an appropriate evaluation index for damage severity.(5)Tea plants attacked by two pest species were detected by E-nose and GC-MS,and the ratio of two pest species was different.The results of GC-MS showed that VOCs emitted by tea plants attacked by two pest species in different ratios were different.SVM was employed for classification and regressison analysis.Good results were obtained,which indicated that E-nose was a feasible technique for pest species ratio prediction.