Near Infrared Spectrum Data Analysis Of Tobacco Based On Deep Learning

In cigarette quality monitoring and production,it is important that accurate determination of cigarette of conventional chemical composition to guarantee the quality of cigarettes.In thesis,we study the prediction model of chemical components of tobacco based on near infrared spectrum.In the thesis,we established spectrum model by the deep learning method,and studied the classification model of tobacco leaf planting area are based on the residual neural network,and the regression model of tobacco leaf chemical components are based on Long Short-Term Memory neural network.1.Classification method of tobacco growing area by near infrared spectrum based on residual neural network: In this chapter,the deep learning method was combined with tobacco leaf NIR spectrum,and the residual neural network was used to extract NIR spectrum features and effectively identify them.A residual neural network classification model based on near-infrared spectroscopy of tobacco leaves was proposed to identify and classify planting areas of tobacco samples.2.Regression model of nicotine in tobacco leaf based on full convolutional neural network: In this chapter,we proposed the 1D-FCN regression model for the quantitative analysis of nicotine in tobacco by near infrared spectroscopy.The deep learning algorithm used to analyze near-infrared spectral data can predict tobacco chemical composition better than traditional machine learning methods.In this model,convolutional layer with a step size of 2 is used to replace the pooling layer to avoid the loss of spectral information in the process of feature extraction.3.Analysis model of tobacco chemical constituents based on Long Short-Term Memory neural network: In this chapter,we proposed the Res Net-LSTM network to quantitative analysis of various chemical components of tobacco based on the research contents of the previous two chapters.In the model,the residual network is used to replace the traditional convolutional neural network for feature extraction,which improves the prediction accuracy of the model.Through Long Short-Term Memory neural network,we can selectively transmit information through the data information to be deleted and stored in the gating unit.