Remote Sensing Monitoring Of Wheat Scab Based On Hyperspectral Data

In this thesis,wheat scab is used as the research object,using the field surveyed imaging hyperspectral data and non-imaging hyperspectral data to study dieases at three scales of wheat grain,spike and canopy.The feature variable is used as the input of the classification algorithm to construct a monitoring model,which provides scientific guidance for the identification and control of wheat scab disease in agricultural production.The main research work is as follows:(1)On the grain scale,use imaging hyperspectroscopy to obtain healthy and diseased wheat grain hyperspectral data as the research object,and the identification model of wheat scab was constructed by screening sensitive characteristic bands.Firstly,the original data is preprocessed by orthogonal signal correction,and secondly,the genetic algorithm combined with partial least squares method is used to screen out 8 sensitive feature bands.Finally,the characteristic band and full-band spectral data were used as the input of the model,and the wheat grain scab recognition models based on support vector machine and random forest were established respectively.Among them,the model based on GA-PLS has better recognition accuracy and model running time than the recognition model based on full band.The recognition accuracy of SVM model is slightly higher than that of RF model,but the model running time is longer.The recognition accuracy is slightly lower than the SVM model,but the model runs for a short time.The SVM model based on GA-PLS has a sample prediction accuracy rate of 100%,and the result is optimal.(2)On the spike scale,imaging hyperspectromspikes were used to collect hyperspectral image data of wheat spikes of different disease severity,and the classification model of wheat scab severity was constructed by combining spectral features and image features.Firstly,wheat spike regions and diseased areas were segmented based on image features and the disease severity was calculated quantitatively.Secondly,the continuous projection algorithm(SPA)was used to screen out 12 sensitive characteristic bands as spectral characteristic variables.For each sensitive band feature image,the gray texture co-occurrence matrix(GLCM)was used to extract the four texture features of contrast,energy,entropy,and correlation.Based on the RGB color model,the three components of each sample image,R,G,and B,were calculated.Based on this color moment feature,based on the correlation analysis method,the extracted texture features and color features are filtered,and the texture features and color features with higher correlation are selected as input variables of the final model.Finally,the particle swarm optimization support vector machine algorithm(PSOSVM)was used to construct wheat scab disease diagnosis models based on different feature variable combinations.Among them,the PSOSVM model based on the combination of spectrum and color features has the best results,and the accuracy of the training set and prediction set is95% and 92%,respectively.The use of hyperspectral images and the combination of image and spectral feature information can accurately and effectively diagnose the severity of wheat scab disease.(3)On the canopy scale,a non-imaging ground feature spectrometer was used to collect healthy and diseased wheat canopy hyperspectral data as research objects.Using random frog jumping(RF),competitive adaptive weighted resampling(CARS),and Variable combination cluster analysis(VCPA)three-variable screening method.Based on particle swarm algorithm optimized support vector machine,different wheat canopy scab disease surveillance models are constructed and compared.The RF algorithm,CARS algorithm,and VCPA algorithm were used to screen 10,9,and 8 sensitive feature bands,respectively.Among them,the monitoring model established based on the VCPA algorithm to screen sensitive bands has the best accuracy,with an accuracy of 92%,which is better than RF(79%)And CARS(88%)based monitoring results.