Identification Methods Of Tobacco Maturity Based On Image Processing

Tobacco field maturity,as an important factor affecting the quality of tobacco leaves,is difficult to be accurately identified by tobacco farmers only with the naked eye.In order to realize the accurate identification of tobacco maturity in the tobacco harvesting link,the following tobacco leaves are taken as samples,they are preprocessed based on image processing technology,and then the corresponding tobacco leaves maturity discrimination models are established based on traditional machine learning(BP neural network and support vector machine)and deep learning(Alex Net and VGG16Net);by analyzing and comparing the classification effects of each model,the optimal model is obtained.The test results show that the method for discriminating the maturity of tobacco leaves based on image processing technology,traditional machine learning and deep learning is feasible,which lays the foundation for the further development of tobacco harvesters.The main research contents and conclusions of this article are as follows:(1)Tobacco leaves maturity identification methods based on traditional machine learning.Pre-process the collected samples,first use the median filtering algorithm to denoise,then use the two-dimensional gamma function uneven illumination image adaptive correction algorithm for lighting balance,and finally use the improved maximum between class variance method and mathematical morphology for segmentation.After preprocessing,based on the color space model and gray level cooccurrence matrix,the color features(R,G,H,S,and V values)and texture features(energy,entropy,moment of inertia,and correlation)of the target image are extracted.Using the extracted features as input and BP neural network as the classification technology,a model of tobacco maturity identification based on BP neural network is obtained by training model in 1208 samples in the training set,and 403 samples in the test set are predicted with an accuracy rate of 90.82%;support vector machine as a classification technology,through the training set of 1208 samples training model,403 samples in the test set for prediction,the accuracy rate is 94.54%.(2)Tobacco leaves maturity identification methods based on deep learning.The affine transformation algorithm was used to enhance the data of 1208 samples in the training set,and the samples in the training set were expanded to 4832.Alex Net was used as a classification technology to obtain a tobacco leaves maturity discrimination model based on AlexNet under different optimizers.Trained under the optimizer SGDM,the obtained model predicted 403 samples in the test set with an accuracy rate of 97.77%;trained under the optimizer RMSProp,the obtained model predicted 403 samples in the test set with an accuracy rate of 96.28%;trained under the optimizer Adam,the obtained model predicts 403 samples in the test set with an accuracy rate of 98.26%.Affine transformation algorithm was used to enhance the data of 1208 samples in the training set.The samples of the training set were expanded to 4832.VGG16 Net was used as the classification technology.The model was trained under the optimizer SGDM to obtain a tobacco leaves maturity identification model based on VGG16 Net.403 samples were used for prediction with an accuracy rate of 98.76%.(3)In the methods of tobacco leaves maturity identification based on traditional machine learning,the model established by using support vector machine as the classification technology has higher accuracy;in the methods of tobacco leaves maturity identification based on deep learning,the model established by using VGG16 Net as the classification technology has higher accuracy;The identification accuracy of deep learning models are generally higher than that of traditional machine learning models.