Remote Sensing Monitoring Of Wheat Yellow Rust Based On Canopy And Regional Scale

This thesis uses remote sensing to monitor wheat yellow rust in different scales.Based on field experiments,canopy spectral data and the severity of wheat yellow rust were obtained.At the same time,relevant data such as Sentinel-2 satellite imagery and weather data were obtained.Monitoring and research on wheat yellow rust based on canopy and regional scales,and using different characteristic variable screening and classification algorithm to establish monitoring model to provide timely and scientific guidance for wheat yellow rust control.The main research work is as follows:(1)On the canopy scale,according to the hyperspectral response spectral characteristics of wheat yellow rust canopy obtained from field experiments at different growth stages,the canopy spectrum changes and response laws of wheat yellow rust were studied.Monitoring and analysis of wheat yellow rust on the canopy scale,provided more theoretical basis for later regional scale research.In order to analyze the correlation between spectral characteristics and the degree of wheat yellow rust,and analyze the characteristics with higher correlation from the original spectrum and vegetation index characteristics,respectively.The K-Means algorithm was used to select the optimal feature set,and the LSSVM classification algorithm was used to establish wheat yellow rust monitoring model in different growth stages.Among all the models,the precision of K-Means and LSSVM monitoring models on May 18 reached maximum of 80.5%,which basically realized the monitoring of wheat yellow rust on the canopy scale.(2)On the regional scale,this paper builds a wheat yellow rust severity monitoring model based on Sentinel-2 remote sensing satellite image data.Through K-Means combined with the ReliefF algorithm,three wide-band exponential features EVI,SIPI,SR,and two red-edge band exponential features NREDI2 and NREDI3 were used as input variables of the model.Two monitoring models of yellow rust were established using BPNN method to monitor the severity of yellow rust in wheat in Ningqiang County,Shaanxi,and the results of the two data models were compared and analyzed.The results show that the monitoring effect of the BPNN model using the wide-band vegetation index combined with the red-edge band vegetation index index as input variables,It is better than the model using only the broad-band exponential feature as the input variable,and its overall accuracy reaches 83.3%,which effectively improves accuracy of the wheat yellow rust severity monitoring model.(3)At the regional scale,based on the theoretical basis of Chapter 4,the coupled meteorological data combined with remote sensing data form multi-source data,with view to establishing a more accurate model of wheat yellow rust severity monitoring.Used the ReliefF and K-Means methods,meteorological and remote sensing characteristic variables has a high degree of correlation with the disease are screened.OVO SVMs algorithm was used to establish three monitoring models(remote sensing data model,meteorological data model,and remote sensing combined with meteorological data model)to monitor the occurrence of wheat yellow rust on regional scale.The results show that the remote sensing meteorological data model has the highest monitoring accuracy,with overall accuracy of 83.3%.This shows that multisource data enables the model to incorporate more effective information related to disease and improve the accuracy of disease monitoring.