Discrimination of cotton plants damaged by cotton bollworm of different instars based on electronic nose use

摘要

The cotton bollworm, Helicoverpa armigera, is one of the three main cotton pests and causes serious damage to the yield and quality of cotton fiber in the world. In this study, an electronic nose (E-nose) was employed to detect cotton plants damaged by cotton bollworm of different instars. Stable value, mean differential value, wavelet energy and the coefficients of the fitted quadratic polynomial function were extracted as four kinds offeature parameters from the electronic nose curves. Principal component analysis (PCA), Multi-layerperceptron (MLP) neural network, Radial basis function (RBF) neural network and Extreme Learning Machine (ELM) were used to analyze the data. The results indicated that the PCA analysis of four feature parameters couldn't discriminate the cotton plants damaged by cotton bollworm of different instars perfectly. The classification accuracies of MLPNN, RBFNN and ELM based on four feature parameters were quite satisfying which were both more than 90%. ELM had the best classification performance that the accuracies of training set and test set based on four feature parameters were all 100%. The wavelet energy value had a good classification performance based on the three neural networks. It indicates that the wavelet energy value is potential in the feature extraction of E-nose data. It could be concluded that E-nose is a potential technique for detecting cotton plants damage by cotton bollworm of different instars.