Remote Sensing Monitoring Of Winter Wheat Scab And Powdery Mildew Based On Different Scale

This study uses remote sensing to detect wheat disease at different scales.Studing on two typical wheat diseases wheat scab and powdery mildew,using hyperspectral data from field surveys and remote sensing image data to detect diseases at spike,canopy and regional scales of wheat.The monitoring model was built using sensitive feature variables and classification algorithms to assist the Ministry of Agriculture in the management of wheat crops.The main work and progress are as follows:(1)On the spike scale,sensitivity feature screening of wheat scab and mo nitoring model construction method was proposed by using the difference in re flectivity caused by changes in water structure and healthy wheat ears.In this chapter,based on the measured of wheat scab and healthy wheat ears spectral data,the reflectance characteristics,vegetation index characteristics and detailed features were combined with correlation analysis,SVM and GASVM classificatio n methods to establish monitoring model of scab.Screening for the sensitivity and significant difference of scab,including three characteristic bands of 350nm-400nm,500-600nm and 720-1 000nm,three vegetation indices of MSR,NDVI and SIPI,and two wavelet features of WF01 and WF02.A total of 16 wheat scab monitoring models were established by combining the selected optimal c haracteristic variables with unoptimized SVM and SVM(GASVM)optimized b y GA algorithm.In all models,the monitoring accuracy of MSR combined wit h GASVM is the highest with an accuracy of 75%.(2)On the canopy scale,this chapter used the study of spike to select the characteristic variables suitable for canopy scale disease monitoring.In this ch apter,the canopy spectral data of wheat scab and healthy samples are determin ed by experiment.Then reflectance characteristics,vegetation index characteristi cs and detail features combined with correlation analysis and SVM and GASV M classification methods established a canopy-based scab monitoring model.Th e characteristics of disease sensitivity and significant difference were screened,including two characteristic bands of 500-600 nm and 700-1000 nm,four veget ation indices of MSR,ARI,TVI and TVInew,and two wavelet features of W F01 and WF02.A total of 16 wheat scab monitoring models were established by combining the selected optimal characteristic variables with unoptimized SV M and SVM(GASVM)optimized by GA algorithm.In all models,TVInew co mbined with GASVM established monitoring model is the highest monitoring a ccuracy with an accuracy of 88%(3)Wheat powdery mildew is one of the most serious diseases in wheat,so timely and effective monitoring of wheat powdery mildew is necessary for i mproving wheat yield and quality.The selection of suitable modeling method a nd feature selection algorithm play an important role in improving the perform ance of remote sensing monitoring of crop diseases.In this study,the GF-1 re mote sensing image was used to extract total fourteen characteristic variables,i ncluding four band reflectance data and ten vegetation indices.An approach co mbining relief and minimum redundancy maximum relevance(mRMR)algorith ms(relief+mRMR)is proposed for improving the ability to remove redundancy of relief algorithm.First,the relief algorithm was used to calculate the weight of each feature and filter out the disease independent features.Then the mRM R algorithm was used to remove the redundant features.Finally,the optimal fe ature set which including NIR,SR and NDVI was as the input variables.Mea nwhile,the other two feature sets such as SR,GNDVI and TVI by relief algo rithm and TVI,RTVI and RDVI by mRMR algorithm were also obtained.Sup port vector machine(SVM)is a machine learning method based on statistical 1 earning theory,its working principle is to minimize the structural risk as the c ore,improve the generalization ability,and resolve problems such as nonlinear,small samples,etc.Penalty factor and nuclear parameter should be considered i n establishing monitoring model for wheat powdery mildew,and traditional par ameter selection was mostly through multiple experiments.Presently,the comm only used grid search(GS)algorithm obtained the optimal parameters,but its e fficiency is low and workload is large.The advantage of the genetic algorithm(GA)is to solve the global optimal problem,robust and could be independent of the domain of the problem when searching quickly.Therefore,SVM optimiz ed by genetic algorithm(GASVM)was used to monitor wheat powdery milde w in Hebei,China.For comparison and validation,SVM method,SVM optimi zed by GS algorithm(GSSVM)approach,and three existing wheat powdery mi ldew monitoring methods such as AdaBoost,particle group optimized least squ are support vector machine(Pso-LSSVM)and random forest(RF)were also be en used.The results illustrated that the performance of models constructed usin g the feature set through relief+mRMR algorithm outperformed the models esta blished using the feature set through only relief algorithm or only mRMR algo rithm.The result demonstrated that the combination of relief and mRMR algori thms can effectively remove the redundancy between features while selecting hi gh disease correlated features.Additionally,in three monitoring models based f eatures selected by relief+mRMR,the relief+mRMR-GASVM monitoring mode 1 was with the highest overall accuracy of 85.7%,and which increased by 21.4%and 7.2%than relief+mRMR-SVM and relief+mRMR-GSSVM monitoring models.The result indicated that the relief+mRMR-GASVM approach can effec tively improve the monitoring accuracy and consistency of the wheat powdery mildew model,and further strengthen the reliability of the model in practical a pplications.Furthermore,the monitoring accuracy of GF-1 data combined with relief+MRMR-GASVM model increased by 21.4%,14.3%and 7.1%than AdaB oost,PSO-LSSVM and RF methods,respectively.These results revealed that th e GF-1 data combined with the relief+mRMR-GASVM model can be used for remote sensing monitoring of wheat powdery mildew.