Quality Classification Of Flue-Cured Tobacco Leaf Based On Convolutional Neural Network(CNN)

China is a large agricultural country.Tobacco leaves are an important agricultural pillar industry and an important supplement for farmers to increase their income(for convenience of description,in this paper,flue-cured tobacco leaves are collectively referred to as tobacco leaves);tobacco leaf classification is the most important part of tobacco production.The rationality of tobacco leaf grading determines the final economic benefits of tobacco farmers.At the same time,with the gradual refinement and standardization of the national standards for tobacco leaf grading,in the daily tobacco leaf purchase work,higher requirements have been placed on how to accurately identify tobacco leaf grading.In the daily tobacco leaf grading process,it mainly relies on the harvesters to classify the tobacco leaves.The level of its business level plays a decisive role in the grading process.In the grading process,most of them are described by fuzzy natural language.Qualitative classification without quantitative indicators is prone to problems such as differences in artificial tobacco leaf grading,unstable grading results,and low pass rate,which in turn leads to a series of livelihood disputes such as waste of tobacco leaf resources and conflicts between tobacco farmers and tobacco leaf purchase stations.China is a large agricultural country.Tobacco leaves are an important agricultural pillar industry and an important supplement for farmers to increase their income(for convenience of description,in this paper,flue-cured tobacco leaves are collectively referred to as tobacco leaves);tobacco leaf classification is the most important part of tobacco production.The rationality of tobacco leaf grading determines the final economic benefits of tobacco farmers.At the same time,with the gradual refinement and standardization of the national standards for tobacco leaf grading,in the daily tobacco leaf purchase work,higher requirements have been placed on how to accurately identify tobacco leaf grading.In the daily tobacco leaf grading process,it mainly relies on the harvesters to classify the tobacco leaves.The level of its business level plays a decisive role in the grading process.In the grading process,most of them are described by fuzzy natural language.Qualitative classification without quantitative indicators is prone to problems such as differences in artificial tobacco leaf grading,unstable grading results,and low pass rate,which in turn leads to a series of livelihood disputes such as waste of tobacco leaf resources and conflicts between tobacco farmers and tobacco leaf purchase stations.In this paper,a large amount of field data is collected on the ground,and the data is carefully analyzed.Based on the characteristics of many types and large amounts of data,a classification algorithm for flue-cured tobacco leaves based on convolutional neural network(CNN)is proposed.The specific work is as follows:1.By collecting tobacco leaf data on the spot,taking photos of the tobacco leaf images under natural light,establishing a tobacco leaf data set Tobacco,and classifying the collected tobacco leaf image data for simulation experiments.2.This article deeply studies the traditional tobacco leaf grading algorithm.In view of the low accuracy of the tobacco leaf grading results and the small number of experimental samples,it comprehensively considers the model parameter amount,model performance,model availability,and the training and computing resource requirements of the model for tobacco leaf data.Factors such as quantity and model ease of use,a tobacco leaf grading algorithm based on convolutional neural networkinception-V3 is proposed,and the inception-V3 network model is reasonably modified.3.The Inception-V3 model cannot achieve the excellent classification effect of tobacco leaf classification in a variety of data sets,such as the tobacco leaf data studied in this paper.In this paper,the Inception V3 model is reconstructed,and the tie pooling layer is used to replace the fully connected layer in the original convolutional neural network.At the same time,in order to reduce the probability of overfitting of the network model,a layer is added after the fully connected layer.The dropout layer further improves the computational efficiency of the neural network to reduce the occurrence of low accuracy of tobacco leaf classification results.In this paper,the use of asymmetric convolution kernels to replace larger convolution kernels can effectively reduce the number of parameters and reduce overfitting,so as to improve the expression capacity of the network's nonlinear expansion model;by increasing the number of output channels,the spatial structure is simplified Transform tobacco leaf data information into higher-order abstract feature information,reduce the input representation of tobacco leaf before spatial aggregation,so that the strong correlation of the results of neighboring units loses less when reducing the dimension;when performing spatial aggregation at a lower dimension,use The multi-branch structure extracts different degrees of higher-order features to achieve the effect of enriching the network's expressive power.At the same time,the network signal is conveniently compressed,making the learning speed faster and effectively reducing the time required for training.Through training and verification experiments on the tobacco leaf data set Tobacco,a good classification recognition effect is obtained,which can effectively improve the accuracy of tobacco leaf recognition.