Hyper Spectral Remote Sensing Monitoring And Spectral Sensitivity Analysis Of Wheat Stripe Rust

Wheat stripe rust is a low-temperature,high-humidity and strong-light fungal epidemic disease caused by Puccinia striiformis f.sp.tritici.It is one of the major types of wheat disease prevention in China.The actual production is generally dominated by undifferentiated regional prevention and control,which increases the cost of wheat planting and pesticide residues in cultivated land.The advancement of agricultural informatization and the development of spectral detection technology provide technical support for rapid,non-destructive,and efficient monitoring of stripe rust.Hyperspectral data can provide detailed surface reflectance spectral information,which can accurately and non-destructively reflect the physical and biochemical components and canopy structure changes of disease-stressed crops.However,the highdimensional and small sample data characteristics of hyperspectral data make their direct application ineffective.Extracting spectral features sensitive to stripe rust in hyperspectral datasets can not only reduce the computational cost but also improve the generalization ability of the model and the model accuracy of remote sensing monitoring of crop diseases.On this basis,the paper took wheat stripe rust as the research object,and integrated canopy hyperspectral data and ground survey data to monitor wheat stripe rust by remote sensing,and the sensitivity of vegetation index to stripe rust at each spectral resolution was calculated using the resampling canopy reflectance data,so as to provide a reference for data dimensionality reduction of hyperspectral data in stripe rust monitoring and the selection of the optimal vegetation index under different spectral resolutions.The major research contents of this dissertation are as follows:(1)In order to reduce the dimension of hyperspectral data and improve the model accuracy of wheat stripe rust monitoring.According to the specific response of wheat plants to stripe rust in the visible-near-infrared spectral range and the reflection of solar-induced Chlorophyll Fluorescence(SIF)on the physiological changes of wheat photosynthesis.The paper selected the vegetation index,"trilateral" parameters,canopy SIF parameters,and characteristic bands that sensitive to stripe rust to form the initial feature set.Then,the max-relevance and minredundancy(mRMR)algorithm and correlation coefficient(CC)analysis were used to reduce the dimensionality of the initial feature set,respectively.Features selected by mRMR and CC were input as independent variables into the extreme gradient boosting regression(XGBoost)and gradient boosting regression tree(GBRT)to monitor the severity of stripe rust.The results of this experiment show that,compared with CC analysis,the monitoring accuracy of the features selected by mRMR in the XGBoost and GBRT models increased by 12%and 17%on average,respectively.Meanwhile,mRMR-XGBoost model achieved the best monitoring accuracy(R~2=0.8894,RMSE=0.1135).The R~2 between the measured DI and predicted DI of mRMR-XGBoost was improved by an average of 5%,12%,and 22%compared with mRMRGBRT,CC-XGBoost,and CC-GBRT models.These results suggested that XGBoost is more suitable to the remote sensing monitoring of wheat strips rust,and mRMR has more advantages than the commonly used CC analysis in feature selection.Field survey data validation results also confirm that the mRMR-XGBoost algorithm has excellent monitoring applicability and scalability.(2)In order to explore the effect of spectral resolution on the monitoring accuracy of wheat stripe rust by remote sensing,this paper simulated canopy hyperspectral data as multispectral data with spectral resolution of 5-80 nm according to the specified spectral interval(5 nm).Thirteen vegetation index values commonly used for stripe rust monitoring were calculated under each band width,and the sensitivity coefficients of each vegetation index to different band widths and the sensitivity coefficient of vegetation index to DI under different band widths were quantified.Based on the optimal fitting model of each index,the remote sensing monitoring model of wheat stripe rust under different band widths was constructed.The experimental results indicated that the normalized difference vegetation index(NDVI)and structural independent pigment index(SIPI)are less affected by the band interference,and they are suitable for sensors of various spectral resolutions to monitor wheat stripe rust.Comparative analysis of the sensitivity of vegetation index to stripe rust disease index found that triangular vegetation index(TVI),plant senescence reflectance index(PSRI),nitrogen reflectance index(NRI),normalized pigment chlorophyll ratio index(NPCI),modified simple ratio index(MSR),red-edge vegetation stress index(RVSI),anthocyanin reflectance index(ARI)and photochemical reflectance index(PRI)have obvious responses to DI in the spectral resolution range of 5～80nm.The comparison experiments on the accuracy of vegetation index remote sensing monitoring DI in different wavelength bands show that the indices MSR,NRI,NDVI,PRI,SIPI and TVI can achieve higher monitoring accuracy in any spectral resolution.(3)To determine the appropriate vegetation index for monitoring stripe rust with different sensors,this paper simulated 6 satellite image data(SPOT-6,GF6-PMS,GF6-WFV,ZY-3,Landsat 8,Sentinel 2)which are commonly used in crop disease monitoring.Based on satellite multispectral data,13 multispectral indices were calculated,and correlation analysis was carried out with the stripe rust disease index,so as to compare the correlation between each multispectral index in different satellite sensors and stripe rust disease.Without considering the effect of spatial resolution,the correlation between each vegetation indices and DI at different satellite sensor levels was not significantly different.The accuracy comparison of hyperspectral data and simulated multispectral data for monitoring the severity of wheat stripe rust in the same study area showed that the sensitivity to wheat stripe rust based on NDVI,NPCI and NRI was less than moderate and severe in mild disease,while TVI was more sensitive to mild stripe rust stress.