Recognition Of Rice Disease Based On Convolutional Neural Network

As one of the most important foods in my country,rice occupies an important position in China’s agricultural production.It is one of the most important tasks to ensure its annual output.Rice disease is one of the main factors affecting rice yield,so it is very important to quickly and accurately identify and detect rice disease and carry out effective control.There are many types of rice diseases currently known to humans,but most scholars currently only focus on a certain disease or several diseases,and traditional image technology and machine learning are strongly dependent on specific features of the image.Although there have been good results in rice damage recognition,the recognition and classification model obtained has certain limitations.Convolutional neural network can just reduce its limitations.It can automatically extract image features through a large number of training samples.In response to the above issues,the research contents of this article are the following issues:1.Combined with image segmentation and convolutional neural network,an image recognition model was designed for four disease images of bacterial leaf blight,red blight,flax spot and sheath blight in the natural environment of rice.Data enhancement methods such as image random rotation and random brightness changes were used to expand the disease data set,then the data were randomly divided into training data and test data at a ratio of 4:1.The three algorithms of region growth,level set-based CV model,and saliency detection were applied to segment the expanded image,and the network layer was constructed with 6 layers(input layer 32×32×3,convolution kernel size 5×5)and 8 layers(input layer 227×227×3,volume Convolutional neural network with core size 11×11,5×5,3×3).The experimental results show that the model obtained by combining the saliency detection segmentation algorithm and the 8-layer convolutional neural network has the best effect.Its training and recognition accuracy rate is 99.99%,and the test recognition accuracy rate is 99.88%,which can accurately detect diseases.Recognition.2.Based on the previous experiment,the end-to-end rice disease image recognition was realized by using Alex Net network with a total of 8 network layers,and the number of rice disease images was increased.Convolutional neural networks were used to identify rice leaf disease images under 8 simple backgrounds,including rice stem nematode disease,bacterial leaf blight,and bacterial leaf streak.After expanding the sample of diseased images by random rotation,random brightness and random contrast,80% of the images are randomly divided into training samples of the convolutional neural network,and 20% are test data.The training samples are directly input into the Alex Net network and Le Net5 network for training,and two models of disease recognition are obtained.On the Alex Net network,the fuzzy C-means clustering image processing and the batch normalization layer are added after each activation function to recognize the image,and two other rice disease recognition models are obtained.Combined with the recognition results of the four models and the analysis of model performance evaluation indicators,it shows that the recognition effect of the model obtained by adding the batch normalization layer to the Alex Net network is the best.The final test recognition rate reached 99.11%,which is 0.23%,0.59% and 4.43%higher than the other three models respectively.The model has strong recognition ability and generalization ability,and provides reference for rice disease research based on convolutional neural network.3.From the above two tests,it can be seen that the convolutional neural network with a deeper network layer has a better recognition effect.Therefore,the simple Lenet5 network is improved in this experiment to enable the network to extract deeper features.At the same time,in order to solve the problem of small sample data,the features identified by the classifier are added by the fusion algorithm.Using the Concat fusion method,a rice disease recognition model combining color features and a two-branch network of improved Le Net5 network is proposed.In order to extract deeper feature images,the Le Net5 network is improved,and in order to obtain more feature images,a two-branch convolutional neural network is designed.The above eight types of expanded rice disease images are taken as data sets and input into the double-branch convolutional neural network to obtain a model.Then the images are converted into an HSV images,and the original images are used as the input of each channel of the two-branch convolutional neural network.After Concat fusion,they are input into the fully connected layer,and finally the recognition result is obtained through the classifier.After comparative analysis,it is found that the recognition model obtained by fusing color features and improving the two-branch network of Le Net5 network has the best effect,which lays the foundation for the future research of fusion feature convolutional neural network in crop disease recognition.Research experiments show that the combination of a convolutional neural network with an eight-layer network and saliency detection can improve the recognition accuracy of four rice leaf diseases in a complex background,and the recognition accuracy can reach99.88%.For the eight kinds of rice leaf disease images in a simple background,the Alex Net network added batch normalization layer and the two-branch convolutional neural network based on feature fusion and improved Le Net network have good recognition results,and the accuracy rates are respectively 99.11% and 99.11%.97.41%,and the network model avoids the over-fitting problem and has strong robustness.