Identification Of Tea Leaves Disease Based On Multi-feature Optimization And Improved Relation Network

As one of the important crops in China,tea is often infected by diseases,especially the diseases of tea leaves.The accurate of tea leaves disease identification can provide policy basis for prevention,reduce the use of pesticides,reduce soil pollution from residual pesticides and prevent the occurrence of large-scale disasters,while increasing the income of tea farmers and enhancing market competitiveness.Based on the research of tea leaves disease and other crop diseases at home and abroad,a tea leaves disease data set was created,and two kinds of tea leaves disease identification algorithms were proposed,which are based on multi-feature optimization of tea leaves disease identification and improved relation network for tea leaves disease identification.The specific work is as follows:(1)Building a tea leaves disease data set.Collect tea leaf images through cameras and UAV(unmanned aerial vehicle)to create tea leaves disease identification data set,includes two subsets of data.Among them,a data set contains three types of tea leaf types: normal tea leaves,leaves infected with tea red leaf spot and leaves infected with tea red scab,totaling100 images.Anther data set contains five types of tea leaf types,such as tea sooty mould,tea anthracnose and so on,with a total of 209 images,which are used to test the ability of tea leaves disease identification algorithms.(2)A tea leaves disease identification algorithm based on multi-feature optimization was proposed.Images of normal tea leaves,tea leaves infected with tea red leaf spot,and leaves infected with tea red scab were studied.First,the tea leaf image's features were extracted using the Histogram of Oriented Gradient and the Inception v3 model.Then,multi-feature optimization processing is performed on two types of extracted features.Finally,the Gradient Boosting Decision Tree algorithm is used as the classifier for the identification of tea leaves diseases.Experiments demonstrate that the multi-feature optimization algorithm reduces the image feature vector from 36068 to less than 150 dimensions while maintaining the high identification accuracy of tea leaves disease,which greatly reduces the complexity of the identification algorithm.At the same time,the identification accuracy of tea leaves diseasebased on multi-feature optimization algorithm could reach more than 95%.(3)An improved relation network for tea leaves disease identification algorithm was proposed.Based on the relation network,the improved relation network in this thesis uses three sub-networks of Inception block,MRFB and MAC to design a few-shot learning framework with high accuracy,robustness and versatility.The Inception block is used to improve the feature expression ability of the embedding model,the MRFB and MAC were used to improve the measurement ability of the relation model.Experiments were conducted on Omniglot and miniImageNet,show that improved relation network greatly improves the accuracy of few-shot learning.Proved the effectiveness of our method.Training the improved relation network on the open crop disease detection data set,trained the network to test the tea leaves disease data set.Which can avoid the disadvantages of small data set of tea leaves.