| Wheat stripe rust is a kind of significant fungal airborne disease affecting high quality and high yield of wheat in China,with the characteristics of wide distribution,strong epidemic and serious harm.It is of great significance to strengthen the early detection and forecast of wheat stripe rust for reducing economic losses,controlling pesticide abuse and ensuring food security.This paper aims to provide new ideas,new methods and new ways for early diagnosis and prediction of wheat stripe rust by exploring the pathogenesis of wheat stripe rust,the spectral characteristics and the propagation law of the diseases.To achieve the desired results,spectral analysis technology,image processing technology,aerodynamics theory,machine learning,and other integrated approach of agricultural information and traditional plant protection technology were utilized to carry out the following topics:early and rapid disease identification and severity assessment of wheat stripe rust based on thermal infrared and hyperspectral techniques;spatial and temporal distribution of stripe rust spores on the basis of Arc GIS and hybrid single-particle Lagrangian integrated trajectory model;intelligent prediction and forecast techniques based on field habitat information of wheat.The main contents and conclusions of the paper are as follows:(1)To achieve a rapid and non-destructive examination of wheat leaves in the incubation stage(infected by stripe rust but not yet conspicuous),a method based on thermal infrared imaging technology was proposed.Thermal infrared images of wheat leaves were continuously collected for 16 days to investigate changes in response of the wheat leaf temperature in the early stages of spore infection.The result showed that the infected wheat leaves could be distinguished from the healthy wheat leaves on the sixth days past inoculation using the proposed method,which was four days earlier than visual observation at the very least.Further statistical analysis showed that with the passage of time past inoculation,there is no significant change in the average temperature and the maximum temperature difference of healthy wheat leaves.On the contrary,the average leaf temperature of inoculated wheat leaves decreased gradually and the maximum temperature difference tended to be increase gradually.Especially at 16 days past inoculation,the average temperature of inoculated wheat leaves was 2.22℃lower than that of healthy leaves and the maximum temperature difference is 1.8℃higher than that of healthy leaves.(2)Traditionally,the severity of wheat stripe rust was mainly determined by the experience and visual grading of plant protection experts,which was high labour intensity and inefficiency.In order to compensate for these shortcomings,this paper proposed a quantitative evaluation method for wheat stripe rust severity using thermal infrared imaging technology.The proposed method firstly effectively enhanced the thermal infrared images of wheat stripe rust leaves with different severity.Then,the Otsu’s method and the temperature difference threshold method were applied to extract the infected region.Finally,the severity of wheat stripe rust disease was evaluated by the proportion of stripe rust area to total leaf area.To verify the feasibility,64 infected leaves with different severity were used to analyze the correlation between the model outputs and the results observed by the plant protection expert.As a result,the correlation coefficients between the expert’s evaluation result and the two models’outputs were both above 0.97.It means that the proposed method was capable to extract the infected region and could be used to evaluate and analyze the severity of wheat.It provided a new method for evaluating the severity of wheat diseases.(3)For early detection of the wheat stripe rust at an early stage,a visual detection method based on hyperspectral imaging was proposed in this paper.Hyperspectral images of wheat leaves infected by stripe rust for 15 consecutive days were collected,and their corresponding chlorophyll contents(SPAD value)were measured using a handheld SPAD-502 chlorophyll meter.Then,the spectral reflectance was extracted from the hyperspectral images,using an image segmentation algorithm based on the leaf mask.The effective wavebands were selected by the loadings of principal component analysis(PCA-loadings)and the successive projections algorithm(SPA).Next,the regression model of the SPAD values in wheat leaves was established based on the back propagation neural network(BPNN),in which the full spectra and the selected effective wavelengths were used as inputs,respectively.The result shows that the PCA-loadings-BPNN model has the best performance,in which the R_C~2,the R_P~2 and the RPD was 0.921,0.918 and 3.363,respectively.Due to the number of effective wavelengths extracted by this model accounted for only 3.1%of the total number of wavelengths,the models were simplified and the computational complexity was reduced significantly.Finally,the optimal model was used to estimate the SPAD of each pixel within the wheat leaf images,and to generate spatial distribution maps of chlorophyll content.The visualized distribution map showed that wheat leaves infected by stripe rust could be identified six days after inoculation and at least three days before the appearance of visible symptoms,which provided a new method for the early detection of wheat stripe rust.(4)Since the wheat stripe rust and the powdery mildew are often mixed in the field and are not easily distinguished,making it difficult to apply for proper medicines according to the disease sources and the etiopathogenesis.To solve this problem,this paper proposed a method for diagnosis and identification of different wheat diseases using hyperspectral imaging.The dataset included 320 hyperspectral samples of powdery mildew,stripe rust,and normal leaves collected by HyperSIS hyperspectral imaging system.The sensitive bands and effective wavelengths for the different diseases were selected by using principal component analysis,successive projections algorithm,and competitive adaptive reweighted sampling,respectively.Then,the classification methods of least-squares support vector machine(LS-SVM)and extreme leaning machine(ELM)were employed to identify powdery mildew,stripe rust and normal leaves.The accuracy rate of all these models were above 94.58%both in the validation set and the testing set.Among these models,the ELM classification model combined with PCA-loadings was the best one,the prediction accuracy of the validation set and the testing set were 99.58%and 100%,respectively.Furthermore,the structure of the model was simple,which only contained three bands,namely 560nm,680nm and 758nm.It provided a theoretical basis for developing a multispectral system for real-time identification of wheat diseases.(5)Wheat stripe rust is a kind of regional epidemic disease,and the airflow propagation of the pathogenic spores leads to large-scale outbreak and epidemic of wheat stripe rust.Therefore,a prediction model of wheat stripe rust based on temporal and spatial dynamic distribution of urediniospores was proposed in this paper.Based on Arc GIS and global re-analysis data,a hybrid single-particle Lagrangian integrated trajectory model was established to simulate the propagation track and settlement density of urediniospores in large area and long time.The influence range in the process of long-distance transmission was analyzed as well.The result showed that the transmission of urediniospores presents in two ways:local transmission and inter-provincial transmission,which meant that the local iterative evolution and the cross-diffusion between provinces and cities existed at the same time.This result provided evidence of upper air current for the annual infection cycle theory of wheat stripe rust in China.At the same time,the single transmission track and sedimentation analysis of wheat stripe rust provided theoretical support for the short-term prediction of wheat stripe rust when a sudden change in the weather occurred,namely to quickly simulate and predict the spread trend of pathogen spores.Analysis of multiple propagation and superposed settlement in large scale and long time provided a new idea for medium and long-term prediction of remote transmission of wheat stripe rust in airborne.(6)Most existing models were dominated by the mathematical-statistical analysis which had poor prediction stability.To solve this problem,intelligent optimal prediction models of wheat stripe rust by habitat information was proposed in this paper,namely back propagation network and support vector machines,respectively.The models were established based on the correlation analysis and effective dimension reduction of wheat habitat information.Then,the parameters and the structure of models were optimized by the genetic algorithm and the particle swarm optimization algorithm,finalized as GA-BP and PSO-SVM.The results showed that the prediction accuracy in the trainning set of the two models both was 100%.The PSO-SVM model was superior to GA-BP model in the prediction accuracy in testing set with shorter average time consuming,indicating that the prediction model of wheat stripe rust based on SR-PSO-SVM algorithm has more advantages in prediction accuracy and running speed.It provided a scientific basis for the medium and long-term prediction of the epidemic degree of wheat stripe rust. |
