Identification Of Crop Leaf Diseases Based On Non-negative Matrix Factorization

Agriculture is the basic supporting industry of Chinese economy,whic h plays a very important role in human life.The production and yield of crops per unit area is closely related to people's life.With the rapid d evelopment of agricultural economy,the traditional disease classificatio n and identification methods have been unable to effectively solve a larg e number of disease problems,intelligent disease classification and iden tification methods are becoming more and more important.With the rapid d evelopment of the Internet era,people can easily collect pictures of dis eases and insect pests.But,In the process of collection,abnormal condit ions such as Gaussian noise,salt and pepper noise or block occlusion app ear in the collected disease and insect pest data set due to weather,sen sor damage or human factors.In this paper,the problem of disease data and disease grade classifi cation with noise is studied based on non-negative matrix decomposition a nd convolution neural network theory.Non-negative matrix decomposition i s mainly used to extract the features of crop leaf disease data,and conv olution neural network is used to process and identify the extracted feat ures.The specific research contents are as follows:(1)Based on the Manhattan matrix decomposition framework,this paper proposes a weighted Manhattan non-negative matrix decomposition method to reduce the impact of noise on the eigenspace by(WNMF).WNMF can not onl y repair the disease data polluted by noise,but also use the repaired da ta to learn more robust feature representation.(2)Based on the sparse coding framework,a feature extraction method based on sparse non-negative matrix decomposition is proposed,which make s the feature space sparse and robust.(3)Based on the theory of non-negative matrix decomposition and convo lution neural network,a disease recognition algorithm model based on con volution neural network is designed.After a large number of data sets tr aining,the feature representation ability of each layer will be graduall y enhanced.High-level features can be obtained at a deeper level,and ha ve better image representation and generalization ability.In this paper,by using the extremely heavy Manhattan non-negative ma trix decomposition method and the sparse non-negative matrix decompositio n method combined with the convolution neural network model,the leaf dis eases of clothing crops can be identified more accurately and effectively,which improves the accuracy of disease identification and reflects the n ecessity of intelligent identification for modern clothing industry.And i mportance.