Research On The Methods Of Wheat Yellow Rust Monitoring Based On Imaging Remote Sensing

Yellow rust is one of the most harmful diseases in wheat production.In recent years,the occurrence of wheat yellow rust has become more and more serious in the context of global climate change.An effective monitoring method for wheat yellow rust is urgently needed to provide a basis for disease prevention and control.The rapid development of remote sensing technology provides a favorable opportunity for the realization of non-destructive,real-time,accurate,and large-scale yellow rust monitoring.However,the current yellow rust monitoring research based on remote sensing is not sufficient to excavate disease information represented by the image and spectral features of remote sensing images,and the yellow rust monitoring models and algorithms at different scales are not perfect.Therefore,this study took wheat yellow rust as the research object,based on the occurrence and development mechanism of yellow rust,used multi-source imaging remote sensing data to carry out the research on wheat yellow rust monitoring methods at leaf,canopy and regional scales.A method of fusion image and spectral features suitable for monitoring of wheat yellow rust at different scales was proposed;Established an early monitoring model for yellow rust based on UAV hyperspectral images,and evaluated the impact of spatial scale on the accuracy of yellow rust monitoring;On the regional scale,a disease monitoring model with high accuracy was constructed based on the integrated multi-features of satellite images;The existing remote sensing monitoring level of yellow rust has been improved from different scales,which provided a certain basis and support for wheat yellow rust prevention and control.The main research results are as follows:(1)At the leaf scale,disease monitoring based on image and spectral features was carried out on the basis of ground imaging hyperspectral remote sensing data.Firstly,the spectral response characteristics of wheat yellow rust were analyzed,and it was found that the spectral reflectance of yellow rust was higher in the visible region(550-700 nm)than that of healthy leaves,and lower in the near infrared region(730-1000nm)than that of healthy leaves.The spectral reflectance of leaves with different disease severity decreases with the increase of disease severity in the near-infrared region.Secondly,SPA and CFS algorithms were used to select the original spectral bands,vegetation indices and texture features sensitive to disease.A leaf lesion extraction model and a leaf disease severity estimation model based on the fusion of spectral and texture features were constructed with sensitive features.The results showed that the accuracy of the model based on the fusion of spectral and texture features was better than the model based on spectral or texture features.For the model of leaf lesion extraction,the SVM model based on the fusion of vegetation indices and texture features obtained the best performance(95.8%),which was 6.3%higher than the model based on vegetation indices.For the estimation model of leaf disease severity,the R~2 of the PLSR model based on the combined vegetation indices and texture features was0.92,which was 0.05 higher than that of the PLSR model based on the vegetation indices.In general,the proposed model of wheat yellow rust leaf lesion extraction and leaf disease severity estimation based on fusion of spectral and texture features have effectively improved the model accuracy,which provided a new idea for wheat yellow rust monitoring.(2)At the canopy scale,UAV hyperspectral images were used as data sources to monitor wheat yellow rust.Firstly,the spectral response mechanism of wheat yellow rust at canopy scale was explored.It was found that the spectral response of yellow rust at canopy scale was different from that at leaf scale.The spectral reflectance of yellow rust at canopy scale varied greatly in the near infrared region,while the spectral reflectance of yellow rust at leaf scale varied greatly in the visible region.The spectral differences between diseased and healthy wheat increased with the increase of infection period.The physicochemical parameters of wheat infected by yellow rust were significantly correlated with the severity of the disease.Among them,the chlorophyll and nitrogen balance index have a very significant negative correlation with the disease index(R~2:0.66 and 0.73),and the dry matter content and anthocyanins were significantly positively correlated with the disease index(R~2:0.81 and 0.62).Secondly,the effectiveness of the proposed method of fusing spectral and texture features at the leaf scale for monitoring yellow rust at the canopy scale was verified.The sensitive vegetation indices and texture features under different infection stages and different spatial resolutions were extracted and selected,and the two features were merged to establish a yellow rust monitoring model(VI-TF)based on PLSR.The results showed that the VI-TF model also has a good performance on the canopy scale.For example,in the later stage of infection,the R~2 of the VI-TF model was 0.80,which was 0.1 higher than the VI model.Thirdly,the impact of the spatial resolution of the UAV image on the monitoring accuracy was evaluated,and it was found that the spatial resolution has a greater impact on the TF model,but has a smaller impact on the VI and VI-TF models.And it is concluded that the UAV image spatial resolution of 10 cm has a great advantage in monitoring wheat yellow rust.Finally,a method for early detection of yellow rust was proposed by fusing wavelet features and vegetation indices(WF-VI).The results showed that this method has better performance for early detection of yellow rust.In the asymptomatic period and the early stage of symptom,the disease monitoring accuracy of the SVM-based WF-VI model were 75%and 79.2%,which were 12.5%and 10.4%higher than that of the VI model,respectively.In general,the yellow rust monitoring method that combined spectral and texture features also has better performance at the canopy scale;The spatial resolution of UAV images had a greater impact on the TF model;The optimal scale for monitoring yellow rust based on UAV images was 10 cm;A model for early detection of yellow rust based on UAV hyperspectral was proposed,which provides a basis for early detection of yellow rust.(3)At the regional scale,based on leaf and canopy scale studies,multi-spectral satellite images(Planet and Sentinel-2)were used as data sources and a variety of machine learning algorithms were combined to construct a regional scale yellow rust monitoring model.Firstly,three texture indices(DTI,RTI and NDTI)were constructed based on high-resolution Planet images to solve the problem that the ability of original texture features to represent diseases is weakened at the regional scale.The results showed that the yellow rust monitoring ability of the three texture indices was better than that of the original texture features,among which NDTI performed the best.In addition,a monitoring model(TI-VI)combining NDTI and vegetation indices was constructed.The results showed that the SVM-based TI-VI model achieved the highest monitoring accuracy(90.5%),which was 5.1%higher than the VI model.Secondly,on a larger regional scale,based on Sentinel-2 images,a method for monitoring yellow rust that integrates multiple features(spectral,texture,and temporal features)was proposed.A modified bi-temporal band ratio(MBTBR)was constructed to characterize the development of the disease.It was proved that this index had a better performance than the temporal feature(NTVI)existing in previous studies.In addition,the yellow rust monitoring model based on multiple features has higher monitoring accuracy than the model based on spectral features or the fusion of spectral and texture features.For example,in the disease monitoring on April 18,the accuracy of the multi-feature-based SVM model was 86.3%,which was 9.8%and 3.9%higher than that of the model based on vegetation indices and the model combining vegetation indices and texture index,respectively.In general,Planet satellite images have great potential in yellow rust monitoring,and the proposed texture index can effectively improve the ability of original texture features to characterize the disease at regional scale.The MBTBR constructed based on Sentinel-2 images can better characterize the development of yellow rust.The proposed method for monitoring yellow rust disease at regional scale with multi-feature fusion has a better performance,which lays a foundation for monitoring crop disease at regional scale.