Monitoring Of Wheat Scab Based On Multi-source Remote Sensing Data

Since ancient times,China has been agriculture oriented,and the quality and yield of wheat need to be focused.Through remote sensing technology,the thesis is to effectively monitor wheat head blight.Investigation by professional relevant personnel is the traditional monitoring method for wheat scab.It is not only inefficient and wastes resources,but also damages the normal growth of wheat in the field.Now with the means of remote sensing,these problems can be effectively solved.Satellite and UAV remote sensing can continuously and dynamically monitor crop diseases and insect pests.The thesis conducts research from these two aspects.Satellite images use Sentinel-2 images for winter wheat planting area extraction and scab monitoring,and use drone multispectral and hyperspectral to establish winter wheat scab monitoring models.The main research contents are as follows:(1)Monitoring of wheat scab based on UAV imageUsing UAV hyperspectral images in the two periods of the test field to obtain the original spectral characteristics,vegetation index and dual-phase vegetation index,and through correlation analysis to obtain sensitive characteristics that are highly correlated with winter wheat scab.The sensitive features with high correlation with the infection of winter wheat are obtained through correlation.Taking these characteristics as input,the three classification algorithms of SVM,RF and ETC were used to establish the monitoring model of winter wheat scab in mild to severe disease.The results show that the overall accuracy of RF model of UAV hyperspectral image is 94%,which is better than SVM and ETC,2% and 3% higher than SVM and ETC.the kappa coefficient of RF classification model is also the best.Due to the limitation of the number of bands in UAV multi-spectral images,some vegetation indices cannot be used for scab monitoring.The original waveband,vegetation index and texture features are used as initial features,and sensitive features are obtained through correlation analysis.Using the same classification algorithm as the hyperspectral image to establish a wheat head blight monitoring model,the results show that the overall accuracy of the RF model is 87%,which is 3% and 5% higher than SVM and ETC.And it can be found that the accuracy of UAV hyperspectral image classification results is better than that of multispectral,possibly because the number of hyperspectral image bands is more,and the selected features are more closely related to the severity of wheat disease.(2)Winter wheat planting area extraction and scab monitoring based on sentinel-2 satellite imageThrough some field surveys to obtain ground sample data,the study mainly selected five types of ground objects,including water,buildings,wheat,non wheat vegetation and bare land.Then the common spectral features are selected and the sensitive features are screened by correlation analysis.Firstly,all the feature variables were used as inputs to establish the winter wheat classification models of MAD,MID,SVM,PL and SAM,and the models of MAD,SVM and SAM with higher overall accuracy were obtained.Then,the optimized feature set was used as input to obtain the classification results and accuracy evaluation.The accuracy of classification results shows that after feature selection,the overall classification accuracy of MAD,SVM and SAM models is improved by 2.30%,5.30% and 5.30% respectively.The overall accuracy and kappa coefficient of mad model are the best among the three models.Combined with some methods and results of area extraction,this chapter also uses a sentinel 2 image data which is consistent with the time of sampling points on the ground to carry out correlation analysis and independent sample t-test on the selected spectral features such as original bands and vegetation index T-test),which has a strong correlation with the health status of winter wheat and has significant differences between healthy and infected samples,was selected to establish a two classification model of healthy and infected winter wheat.Three methods,MAD,SAM and SVM,were used to establish the remote sensing image monitoring model of winter wheat scab.From the accuracy of the classification results,MAD,SAM and SVM have good accuracy for winter wheat scab,which shows that it is effective to use sentry 2 image for winter wheat scab monitoring.Among them,the overall accuracy of MAD,SAM and SVM models were 86.1%,80.6% and 75.0%,respectively.The overall accuracy of MAD model was the highest among the three monitoring models,which was 5.5%and 11.1% higher than SAM and SVM models,respectively.It showed that MAD model was more suitable for monitoring winter wheat scab with satellite remote sensing images.