Research On Rice Stripe Disease Monitoring Method Based On Multi-Source Data

Rice bacterial streak disease is one of the main diseases that harm the yield and quality of rice in China.It is of great significance to strengthen the monitoring and early warning of thin streak disease for the improvement of disease defense and food safety production.The traditional manual methods of monitoring crop diseases and insect pests mainly rely on human identification,which is inefficient,subjective,inaccurate and low currency.Remote sensing technology has the advantages of large scope,real time and easy data acquisition,which makes it of great practical value and research significance in monitoring crop diseases and insect pests.The traditional remote sensing image classification method has a high degree of artificial participation,and it is difficult to effectively use the spatial information and spectral features in the images.As an emerging technology,convolutional neural network can extract refined information of image data and obtain more accurate classification results by virtue of its multi-layer convolutional network features.In this dissertation,based on the deep learning theory,the convolutional neural network model was used to classify the prevalence levels of rice stripe disease after the acquisition of multi-source image data of rice stripe disease in the study area.The main work of the research is as follows:(1)Using camera RGB data,unmanned aerial vehicle(UAV)high resolution image data,drones multi-spectral data and vegetation index NDVI data produced rice thin disease illness for convolution neural network classification grade sample data sets,combined with field survey data will be divided the sample data set into three categories,i.e.health,mild illness and severe illness.Data enhancement method is used to expand the training data,and the classification performance of the network model under different sample data sets is analyzed.(2)Using Python language and Tensor Flow framework,five convolutional neural network models based on rice thin line disease grade monitoring were built,namely VGG16,Inception＿V3,Inception＿V4,Resnet＿V1＿50 and Inception＿Resnet＿V2networks.Channel parameters of the model were adjusted according to the number of channels of input image data,as well as optimization of network parameters.(3)Classification experiment and result analyses of rice thin streak disease based on convolutional neural network.In this dissertation,Inception＿Resnet＿V2convolutional neural network was used to study the monitoring of rice stripe disease,and four other popular methods were established to classify four types of rice stripe disease sample data sets,and the classification results were compared and analyzed.The results show that,among all the classification results,the combination of UAV high-resolution image data and Inception＿Resnet＿v2 network model has the highest classification accuracy,reaching 0.9467.In addition,the Inception＿Resnet＿v2 model has excellent performance in the classification results of the four data types,indicating that residual connection and multi-scale feature extraction are helpful to improve the performance of feature extraction and classification generalization of the model.