Research And Application Of Rice Disease Recognition Method Based On Convolutional Neural Network

Rice is the main grain crop in China.At this stage,with the changes in the global environment,the incidence of rice diseases has increased compared with the past.The traditional method of disease identification is mainly by agricultural experts.At the peak of disease occurrence,there is a situation that is difficult to find an expert.In addition,there is a certain subjective misjudgment in this way,which delays the prevention and treatment of rice diseases,and makes the rice yield greatly reduce or even to crop failures.The situation is causing significant economic losses to growers.In the process of rice growth,the three diseases of rice blast,false smut and bacterial leaf blight have a high infection rate and serious impact on yield.Therefore,this experiment aims to develop a convenient and accurate identification method for these three diseases.In recent years,with the continuous development of machine learning,the convolutional neural network has shown satisfactory results in the field of image recognition.Using a convolutional neural network to solve disease recognition not only promotes the deep integration of artificial intelligence and the agricultural industry but also improves the recognition efficiency of rice disease.This experiment conducted an in-depth study on the identification of rice diseases.First,three types of rice disease sample images were selected for the experiment.Due to the problems of different image size and light intensity,size normalization and histogram equalization were carried out to reduce the impact of these factors.The neural network needs a large number of samples during the training process.In this experiment,the three types of rice disease samples are also enhanced.The methods of data enhancement are mainly random cropping,rotation and mirroring,and the class imbalance is processed for the enhanced data set.Next,the experiment built a 12-layer convolutional neural network model for rice diseases and analyzed the parameters required during the network training process.The initial model has low accuracy in identifying rice diseases,which is difficult to meet the practical requirements.To further optimize the model,this experiment explored the number of iterations,batch size,optimization algorithm,and learning rate,and obtained the optimal parameters through experimental comparison.At the same time,to more clearly understand the learning ability of the model,this study also carried out a visual analysis on the output feature map after the convolution process.The results showed that the recognition accuracy of the optimized model reached 98.24%,and the disease could be accurately identified.Finally,to provide a convenient rice disease identification method,this experiment built a rice disease identification platform based on the Flask framework and designed to implement functions such as uploading disease samples,calling the TensorFlow model and returning disease identification results.When a rice disease occurs,the user can obtain the diagnosis result in time by uploading the disease picture.This makes the growers less dependent on experts and makes it easier for them to carry out further treatment of diseases.Meanwhile,it also provides strong technical support for the identification of rice diseases.