Research On The Recognition Method Of Rice Sheath Blight Based On Deep Learning

Rice is one of the most important food crops in China and plays an important role in national agricultural production.The perennial rice cultivation area in China is about 30 million hectares with an annual output of 200 billion kilograms.There are many kinds of rice diseases,which have a great influence on rice yield.Rice sheath blight is one of the most common diseases in rice cultivation.It is characterized by severe disease,great harm and easy to be ignored by growers.Therefore,how to identify rice sheath blight efficiently and accurately is a problem that the rice growers need to solve urgently.The traditional identification method of rice sheath blight disease in China is mainly judged by plant protection experts with naked eyes.This traditional identification method has long period needed and strong subjective judgment,which is not conducive to the timely prevention and treatment of rice sheath blight disease.With the rapid improvement of computer hardware processing speed and the progress of software technology,the artificial intelligence,image recognition,big data and deep learning technologies have been widely applied in the agricultural field,especially in the field of crop disease diagnosis and recognition.When rice plants encounter diseases,the physiological structure and morphological characteristics of the plants will change,such asdiscoloration,decay,deformation and so on.Therefore,this paper takes the image of Rice Sheath Blight in the northern cold area as the research object,and uses the deep learning technology to identify rice sheath blight.This research play great significance on the improvement of rice yield and the national food security.In this paper,deep learning technology,support vector machine and particle swarm optimization(PSO)algorithm are used to identify images of rice sheath blight.The research content mainly includes the following three parts:(1)Study on image pretreatment of rice sheath blight.Firstly,the median filter algorithm is used to eliminate the noise in the image of rice sheath blight,then Gauss filter is applieded to enhance the image of rice sheath blight,finally,Sobel edge detection operator is used to extract the lesion feature of rice sheath blight,so a feature database of 15000 rice sheath blight images is constructed.(2)Deep belief network model and the DBN that is optimized by particle swarmoptimization model is constructed and trained.A deep belief network with three hidden layer constrained Boltzmann machines is designed and trained.Gradient descent algorithm is used to train the model,and particle swarm optimization algorithm is used to optimize the network structure parameters and learning rate parameters.At the same time,particle swarm optimization algorithm is used to optimize the penalty factor and kernel function parameters in the SVM model.Finally,the optimized deep belief network and the optimized SVM model are used to identify rice sheath blight image and make a comparative analysis.(3)Study on AlexNet model and GoogleNet model for image recognition of rice blight.Based on Python ? TensorFlow environment,AlexNet model and GoogleNet model are constructed to recognizerice sheath blight disease image.In the experiment,when using the traditional shallow neural network learning technology to identify rice sheath blight,the recognition accuracy is low,and the correct recognition rate is only 83.99%.When the advanced deep learning model is used to identify rice sheath blight disease image,the deep belief network shows excellent recognition performance,the correct recognition rate of rice sheath blight achieves 94.05%.After the DBN model optimized by PSO,the correct recognition rate of rice sheath blight is 95.72%.,The Alexnet model and Googlenet model are used to identify rice sheath blight disease image,the correct recognition rates are91.64% and 92.32% respectively.The results show that the deep learning method can effectively identify rice sheath blight and improve the recognition accuracy of rice sheath blight.