Research On The Plant Leaf Recognition Method Based On Modified Convolutional Neural Network

The identification and classification of different plant leaves are very important significance for plant protection and research,ecological environment protection,seed identification,plant pest protection,and so on.Plant leaves are the most important part of plant appearance,which are easy to collect,store for a long time,observe and carry.Owing to computer graphics processing technology,plant leaves have become the main methods and important basis for plant identification and classification.According to plant leaves as the research object,this paper mainly studies three key problems in the field of plant leaves identification and classification,including nonlinear conjugate gradient method,convolutional neural network model,and their applications in plant leaves identification and classification.To improve the theory of a class of nonlinear optimization algorithm represented by nonlinear conjugate gradient method,two improved convolutional neural network models are proposed,investigated and analyzed to realize the high precision,high efficiency and accurate identification and classification of different plant leaves.The details mainly include the following aspects:First,two nonlinear Wei-Yao-Liu(WYL)conjugate gradient methods with fast convergence and strong robustness are proposed,investigated and analyzed for unconstrained optimization problems.To achieve the rapidly convergent ratio of the WYL method,the conjugate gradient parameters of the traditional WYL conjugate gradient method are modified to construct a new search direction.In each iteration,the new search direction satisfies the sufficient descent condition.The global convergence of two modified nonlinear WYL conjugate gradient methods is proved without line search rule.Numerical results show that the modified WYL conjugate gradient method is feasible and effective.Second,for the problem of low identification accuracy and calculation efficiency of leaves of various plants,a convolutional neural network model with high accuracy and fast convergence rate is developed to realize the fast and accurate identification and classification of leaves of different species of plants.The convolution layer is connected with residual error,and a class of Alexnet convolutional neural network based on residual connection is designed and proposed by using conjugate gradient technology to eliminate the disappearance of gradient and gradient explosion.Moreover,in order to accelerate the convergence speed of Alexnet convolutional neural network,we perform batch normalization processing on the input data of the convolutional layer.Combining two different global pooling algorithms can not only reduce the number of feature graphs,but also improve the accuracy and efficiency of algorithm identification.The experimental results show that the modified Alexnet convolutional neural network based on residual connection can efficiently identify different plant leaves.Third,two additional convolutional layers are added to the first convolutional layer due to the above Alexnet convolutional neural network,and then a new Alexnet convolutional neural network is developed,investigated and analyzed by splicing for identification and classification of different plant leaves.By arranging the convolution kernel horizontally and using multiple convolution nuclei of different sizes,different information of plant leaves can be obtained,which can not only identify the category of leaves,but also capture the information of plant leaf diseases,so as to conduct multi-dimensional feature fusion and further improve the experimental effect.At the same time,the widening technique is utilized to avoid the gradient dispersion phenomenon caused by over-deep network.Fourth,the plant leaf recognition system is constructed,which mainly includes the pre-processing technology,image feature extraction and recognition of plant leaf images based on the improved convolutional neural network model.The feasibility and effectiveness of the improved convolutional neural network model proposed in this paper are verified by the plant leaf recognition system.Finally,the main contents of this dissertation are summarized,and further works are discussed.