Research On Monitoring Of Rice Smut Disease Based On Hyperspectral Technology

Food crops are an important guarantee for human survival and life and rice is the most important food for human consumption.However,rice smut has always been one of the important factors restricting its quality and yield.It will appear at any time during the entire period of rice growth,resulting in blackening,corruption,and shedding of rice leaves,cannot growing normally,which eventually led to a reduction in rice production and even loss of grain yield.Under normal circumstances,people will spray pesticides according to the conditions of the disease to prevent the further deterioration of the disease and the proliferation of large areas,which will produce many other problems,such as increased input costs,environmental pollution and so on.Therefore,it is imperative that new methods are needed to solve the problem of rice stress.First,it is necessary to understand the growth conditions of rice in order to take reasonable measures to ensure the normal growth and development of rice.Second,basing on the rice growth information,predict the incidence and seriousness of rice,etc.Thirdly,improving the use efficiency of pesticide has important implications for reducing input costs and inhibiting the spread of rice smut.Hyperspectral technology is a new high-efficiency,non-destructive monitoring technology for crop diseases.It is of great significance for improving rice quality and yield,reducing environmental pollution,and improving sustainable farm management.In this paper,basing on the study of the rice growth mechanism and various classification algorithms under the stress of rice smut,the identification model of healthy and diseased rice can be established,carrying out the accuracy verification of the classification results.Hyperspectral technology provides data support for the early monitoring of rice smut.The main contents are as follows:1)Use hyperspectral devices to collect multiple sets of hyperspectral imagery data from the regions of diseased rice and healthy one.2)Use ENVI image processing software to analyze and process the data collected by the hyperspectral device,and select the ROI region to obtain the spectrum data of the diseased region and the spectral data of the normal region,and use the SVM modeling algorithm to identify it.3)Establish the identification model of rice smut,basing on hyperspectral technology.Use four four kinds of kernel functions of SVM modeling algorithm to identify it,including Linear,Polynomial,Radial Basis Function,and Sigmoid.4)Verify the accuracy of the image classification results.The verification methods include confusion matrix,Kappa statistics.5)The data collected by the hyperspectral device was analyzed and processed using ENVI software,and PCA and ANN methods were used to highlight the affected area.The results showed that after rice is infected with rice smut,and comparing with the spectral curves of the diseased area and the normal area,the spectral reflectance decreased significantly at wavelengths of 450-720 nm and 780-900 nm.This is because in the visible spectrum,Chlorophyll plays an important role in the spectral properties of plants in a great extent.In the blue band centered at 450 nm and the red band centered at 670 nm,chlorophyll strongly absorbs radiant energy and becomes an absorption valley.There is less absorption between the two absorption valleys,forming green reflection peaks and showing green plants.When rice is in the stress of rice smut,yellow-brown lesions appear in the ears and leaves of the rice.The content of chlorophyll in the lesions is reduced,simultaneously the absorption of chlorophyll in the blue and red bands increases,and the reflectance decreases,especially the reflectance of the red band.Therefore,the overall reflectance decreased significantly.In the case of fewer training samples,the classification method of support vector machine has higher diagnostic accuracy.From the results of four kernel functions of SVM identification,we can see that the linear kernel function has the best diagnostic performance and the average classification accuracy is 97.2%.Therefore,the SVM modeling method of linear kernel function is most suitable for the diagnosis of rice diseases affected by rice smut.From the overall classification results,the commission error and omission error scores for the two sets of experimental data were 1.66% and 3.84%,which achieved good results.The study can be used as the basis for the early hyperspectral monitoring of rice smut,and in the future it has practical significance using hyperspectral technology to monitor other diseases and determining the optimal period of prevention and treatment.