Monitoring Of Wheat Yellow Rust Based On Multi-scale Remote Sensing Data

Wheat yellow rust(Puccinia striiformis)is one of the most destrustive disease for wheat and has led to a significant decrease in wheat quality and yield in China.It is very important to identifying and monitoring to guide disease prevention and food safety production.With the developing of various types of remote sensing data,the monitoring of crop disease and management of agricultural through remote sensing is becoming a very important and effective means of obtaining disease information in field condition.The present study took wheta yellow ruts as an example,based on the observation multi-source data,such as the non-imaging hyperspectral,airborne hyperspectral imaging,multi-temporal Sentinel-2 satellite images et al.the corresponding identification and monitoring model of wheat yellow rust is established on the canopy,field and large regional scale,which provides certain theoretical basis and technical support for nondestructive monitoring and effective prevention of wheat yellow rust.The major contents and results in this dissertation are as follows:(1)At the canopy scale,the symptoms of wheat yellow rust were different in different yellow rust infected stages,and the spectral response characteristics of yellow rust in the range of 350-1000nm were studied by obtaining the non-imaging hyperspectral data of wheat canopy in different stages.According to the sensitive spectral characteristic distribution of yellow rust in each stage of disease infected,the wheat stage is divided into early-middle stage(from booting to anthesis)" and middle and late(from filling to milky ripeness).All possible three-band combinations over these sensitive wavebands were calculated as the forms of PRI(Photochemical Reflectance Index)and ARI(Anthocyanin Reflectance Index)at different growth stages and assessed to determine whether they could be used for estimating the severity of yellow rust disease.The optimal spectral index for estimating wheat infected by yellow rust disease was PRI(570,525,705)during the early-mid growth stage with R2 of 0.669,and ARI(860,790,750)during the mid-late growth stage with R2 of 0.888.The classification accuracy for PRI(570,525,705)was 80.6%and the kappa coefficient was 0.61 in early-mid growth stage,and the classification accuracy for ARI(860,790,750)was 91.9%and the kappa coefficient was 0.75 in mid-late growth stage.The classification accuracy of the two indices reached 84.1%and 93.2%in the early-mid and mid-late growth stages in the validated dataset,respectively.The results showed that PRI(570,525,705)and ARI(860,790,750)had strong recognition ability and robustness in the wheat yellow rust discriminant in different yellow rust infected stages at the canopy scale.(2)At the field scale,the low-altitude hyperspectral remote sensing images were used to obtain the imaging hyperspectral images of wheat canopy in the early-mid and mid-late growth stages.The sensitive spectral features of stripe rust were extracted by vegetation index method and continuous wavelet analysis.The sensitive features in the early-mid growth stage contained 6 vegetation indices of GI(Greenness index),MSR(Modified Simple Ratio),NDVI(Normalized difference vegetation index),NRI(Nitrogen reflectance index),PSRI(Plant senescence reflectance index),PRI(570,525,705)and 5 wavelet features of 530nm(6 scale),574nm(5 scale),674nm(4 scale),706nm(1 scale)and 758nm(1 scale),and mid-late stage the sensitive features include 6 vegetation indices of GI,MSR,NDVI,PSRI,TVI,ARI(860,790,750)and 5 wavelet features of 51 Onm(6 scale),662nm(5 scale),714nm(4 scale),766nm(2 scale)and 790nm(3 scale).According to the sensitive features,the exponential inversion model of wheat yellow rust was constructed by PLSR,and the fitting effect of the disease index(DI)inversion model based on wavelet features was higher than that of vegetation index,and the decision coefficient of the inversion model in the early-mid and mid-late stage of R2 was 0.71 and 0.89 respectively,which were higher than the single variable features.Moreover,based on LDA(Linear discriminant analysis)and SVM(Support vector machine)methods,the vegetation index and wavelet feature were used to establish the wheat yellow rust discriminant model at different growth stages.The results show that the SVM model based on wavelet feature has the best classification accuracy of health and yellow rust wheat,the classification accuracy of early-mid stage was 90.7%,and in mid-late stage was 98.7%.(3)At the regional scale,the Sentinel-2 satellite image with high spatial resolution and rich red-edge information was used as the data source to explore the potential of theSentinel-2 satellite for discriminating between yellow rust infection severities(i.e.,healthy,slight,and severe)in winter wheat.The corresponding simulative multispectral bands for the Sentinel-2 sensor were calculated by the sensor's relative spectral response function based on the in situ hyperspectral data acquired at the canopy level.Three Sentinel-2 spectral bands,including B4(Red),B5(Rel),and B7(Re3),were found to be sensitive bands using the random forest method.A new multispectral index,the Red Edge Disease Stress Index(REDSI),which consists of these sensitive bands,was proposed to detect yellow rust infection at different severity levels.The overall identification accuracy for REDSI was 84.1%and the kappa coefficient was 0.76.According to the synchronous Sentinel-2 image and ground survey of Hefei in Anhui Province,the method of optimal threshold was adapted to identify wheat yellow rust at regional scale,and the overall mapping accuracy is 85.2%.(4)At the regional scale,the study conducted the region investigation experiment in Ningqiang county in Shaanxi province in the key growth period of winter wheat,and acquired the synchronous multi-temporal Sentinel-2 images with investigation experiments and the environmental characteristics of the development of wheat yellow rsut.The spectral and habitat characteristics sensitive to the identification of wheat yellow rust were selected by variable projection importance criterion,and the monitoring model of wheat yellow rust with single time,two-temporal and fusion multi-temporal remote sensing data-meteorological data was proposed.The overall classification accuracy of wheat yellow rust monitoring model based on the features of two-temporal normalized vegetation indices(REDSI,VARIgreen,PSRI1,NREDI1,NDVIrel)and meteorological factors(SSD in March,relative humidity in April and May,monthly average precipitation in April and May)reached 84.2%,and kappa coefficient was 0.65.which is better than the two-temporal normalized vegetation indices feature model(the overall classification accuracy is 78.9%)and the single-time phase feature model,which the overall classification accuracy is 73.7%.The results show that the regional scale wheat yellow rust monitoring model of coupled remote sensing and habitat factor makes up for the deficiency of pure vegetation spectral information model,and provides scientific basis for the monitoring and prediction of wheat yellow rust.