| Tobacco mosaic virus(TMV)is a worldwide disease,and its high incidence and incurability are the key obstacles to improve tobacco quality under the lean policy,and early detection and timely prevention and control are important prerequisites to improve tobacco quality.Compared with traditional field survey and sampling methods of disease identification,spectral remote sensing technology has the advantages of being fasted,nondestructive and simple,which provides the possibility of monitoring tobacco mosaic disease over a large area.When tobacco is infected by TMV,its internal physiological indicators will change,resulting in a subsequent change in leaf reflectance at different wavelengths,which provides the principle for spectral monitoring of mosaic disease.Vegetation indices can synthesize crop spectral information and are widely used in crop pest and disease monitoring studies.Based on this,this study used handheld spectrometer and UAV multispectral platform to collect the spectral reflectance and canopy image spectral data of tobacco leaves with mosaic disease,determined the fluorescence parameters of tobacco leaves at different times after inoculation with TMV by chlorophyll fluorescence imaging system,studied the spectral characteristics of different periods and different disease degrees,analyzed the correlation between spectral reflectance,vegetation index and disease degree,screened out the sensitive band of tobacco mosaic disease and the best monitored vegetation index.On this basis,a remote sensing monitoring and recognition model of tobacco mosaic disease was constructed,and the model was applied to the multispectral images of the study area,which provided a new method of the monitoring and identification of tobacco mosaic disease disease.The following research results were obtained:(1)The spectral characteristics of foliar tobacco leave and the spectral characteristics of canopy images of different disease levels were analysed,and it was found that the spectral reflectance of tobacco leaves and the canopy increased to disease levelled in the visible band after foliar disease,while the opposite were true of the near-infrared band.The change in spectral reflectance of tobacco leaves with foliar disease at different times was basically synchronized with the development of the disease,the degree of the disease increased as the fertility of the tobacco progressed;analysis of the first-order differential spectral reflectance of tobacco leaves with different levels of disease severity showed that the position of the red edge shifted toward the blue light,with an obvious"single peak"phenomenon,and the offset gradually increased as the disease level increased.(2)It was found that there was no significant difference in all fluorescence parameter values of control and treated groups before TMV inoculation,and after 6d,12d,and 18d after mosaic virus inoculation,chlorophyll fluorescence parameter values of tobacco leaves changed,Fo’,Fm’,Fv/Fm,and Y(NO)decreased with increasing disease level,while q N parameter values were increased to increasing disease level.(3)The correlation coefficients of leaf spectral reflectance and disease index were calculated to screen out the sensitive wavebands of foliar disease and to construct the corresponding disease estimation and identification models;The random forest and partial least squares regression disease monitoring and identification models were constructed based on the full waveband of the study,and the decision coefficients of both the modelling and validation sets were above 0.9,but the partial least squares regression model had a larger variance compared to the random forest model.(4)Based on the spectral reflectance of UAV multispectral images,a tobacco mosaic disease estimation and identification model was developed using random forest with model y=0.92802x+0.01752,modeling set R~2of 0.96 and RMSE of 3.6507,and validation set R~2of 0.958 and RMSE of 5.5471,with a good linear fit between predicted and measured DI values.(5)Based on the multispectral image data obtained in the study area,four vegetation indices related to disease monitoring were calculated,and through correlation analysis with the mosaic disease disease indices,NDVI and OSAVI were selected as the best disease indicator variables and used to establish a disease estimation and identification model using random forest.The results showed that the determination coefficient of the combined modelling of NDVI and OSAVI reached 0.97 and the RMSE was 0.4,which is much more accurate than the single vegetation index model.The model was applied to the remote sensing image inversion of the monitoring area to derive the tobacco mosaic disease class distribution map,and the accuracy of the remote sensing inversion was tested using 60 sample points in the ground field survey,with a validation result of R~2of 0.8. |
