Tobacco Leaf Mildew Data Analysis And System Design Based On Deep Learning

Tobacco leaf mildew is an important problem facing tobacco industry,and it causes serious economic losses every year.The tobacco leaves in the storage process will cause mildew due to their own moisture content,relative air humidity,storage conditions and other reasons.The internal nutrients of the mildewed tobacco leaves are decomposed and consumed by the mold,and the use value is greatly reduced.In order to find moldy tobacco leaves in a timely and effective manner,so that managers can intervene in advance,thereby reducing the economic loss caused by tobacco moldy,this article builds a one-dimensional convolutional neural network based on the new generation of information technology such as the Internet of Things and deep learning.Based on the network’s tobacco mold status recognition model,a prototype system for intelligent monitoring and early warning of stored tobacco mold has been designed to realize the perceptual monitoring of the environmental parameters of stored tobacco leaves and the early warning of tobacco mold.The main research contents include :1.Designed a tobacco leaf sample data collection test plan,using a self-designed electronic nose device to collect the gas response signal during the moldy process of the tobacco leaf sample,and preprocess the response signal,and at the same time analyze the discrimination of the tobacco leaf samples under different conditions.The results show that tobacco leaf samples in different states have a certain degree of discrimination.Finally,optimize the sensor array to eliminate abnormal and redundant gas sensors.2.In the process of building the model,the main super parameters of one-dimensional convolution neural network model were compared and analyzed,and the optimized super parameters were selected to build the tobacco mildew state recognition model.The experimental results show that the model constructed in this article has a higher prediction accuracy with an average prediction accuracy ratio of100% in the training set and an average prediction accuracy ratio of 89.8% in the test set than BP neural network.Finally,through the feature visualization analysis of the training process of the model,the data classification effect of the model can be displayed intuitively.3.Designed and implemented a prototype system for intelligent monitoring and early warning of the storage of tobacco leaves.Based on the domestically-made innovative GD32 microcontroller and sensor array,a collection terminal for collecting environmental parameters of tobacco leaves was designed,and wireless data transmission was realized through the NB-Io T network.At the same time,an intelligent information management system for the storage of tobacco leaves was built based on Node.js,and the trained tobacco storage moldy status recognition model was loaded through the Flask framework,and finally the intelligent monitoring and early warning of the stored tobacco molds were realized.The prototype system of intelligent monitoring and early warning for the storage of tobacco leaf environment researched and designed in this paper can effectively reduce the occurrence of tobacco leaf mildew and improve the quality of tobacco leaf.Through the comprehensive monitoring of the environmental parameters of the stored tobacco leaves,the intelligent early warning of the mildew of the stored tobacco leaves is realized,and the staff are reminded and guided to deal with it accordingly,thereby reducing economic losses.