Research On Leaf Area Index Inversion And Disease Recognition Of Winter Wheat Based On Remote Sensing Method

As the global warming, the yield loss caused by diseases and pests are increasing in winter wheat. It is necessary for identification of different diseases and guiding for variable rate spraying in wheat. Winter wheat disease mainly includes the yellow rust, powdery mildew, aphids and water stress, etc. Traditional monitoring methods mainly by plant-protecting experts by field investigation, which is hard sledding, time-consuming, poorly timeliness, and often affected by subjective factors of the investors. Remote sensing is able to monitor diseases and inverse of physiological parameters in crops with widespread, all-weather, multiband and lossless characteristics. Firstly, combining support vector machine (SVM) and remote sensing technology, implements an important physiological parameters, winter wheat leaf area index (LAI) of large area inversion. Secondly, as a means of vegetation indices identify different diseases carried out in-depth research. The main research contents and results are as follows:1、Proposed the method using support vector machine regression (SVR) for leaf area index inversion, which could use more band information as input parameters, solved the empirical formula of vegetation index easily saturated and lower LAI inversion accuracy problems. Using the winter wheat’s actual spectra measurement and leaf area index data in the period of erecting stage, elongation stage and filling stage. Chose two vegetation index:the normalized difference vegetation index (NDVI) and the ratio vegetation index (RVI). A total of five kinds of prediction model is constructed, respectively for the two kinds of statistical regression model (NDVI-LAI and RVI-LAI), three kinds of support vector machine regression (SVR) model (NDVI-SVR、RVI-SVR and NRGB-SVR), NRGB-SVR said the input parameters of SVR are blue (B), green (G), red (R) and near infrared (NIR) band. The above five models used the corresponding period environment HJ-CCD data for validation respectively. The results showed that:the root mean square error (RMSE) of NDVI-LAI and RVI-LAI were0.98and0.97; the inversion precision were59.2%and59.3%respectively. As a tool to support vector machines, the root mean square error (RMSE) of the other three models NDVI-SVR、RVI-SVR and NRGB-SVR were0.71,0.83and0.42; the inversion precision were70.4%,67.1%and81.7%.2、Proposed the use of a combination of vegetation indices used to build the feature spacemethods to identify and distinguish the wheat stripe rust, powdery mildew and fertilizerstress, solved using canopy data it is difficult to identify winter wheat conventional stress problem. Fifteen commonly used vegetation indices were selected, and independent t-test was used to get sensitivity index of each stress. Finally, index combination was selected to distinguish the different stresses. The results showed that the combination index (NDVI-PhRI) with normalized vegetation index (NDVI) and physiological reflectance index (PhRI) could be used to identify powdery mildew and yellow rust (PM-YR). A2-dimensional spatial coordinate was established based on NDVI and PhRI derived from hyperspectral data, the different stresses data were displayed in the spatial coordinate, and the classification boundary could be used to identify the powdery mildew and yellow rust stress. Similarly, the combination index (MSR-SIPI) with modified simple ratio (MSR) and physiological reflectance index (PhRI) was used to identify yellow rust and fertilizer-water stress (YR-nOwO); the combination index (NRI-RVSI) with nitrogen reflectance index (NRI) and red-edge vegetation stress index (RVSI) was used to identify powdery mildew and fertilizer-water stress (PM-nOwO). The verification accuracy of PM-YR, YR-nOwO and PM-nOwO models was83.3%,88%,88.75%, the kappa accuracy was63.41%,74.79%,71.43%.3、Put forward a new optimal spectral index is used to identify different winter wheat diseases, solves the problems of limited index to identify the different diseases cannot achieve the ideal effect. The new optimized spectral indices comprised weighted combination of a single band and a normalized wavelength difference of two bands. The most relevant and irrelevant wavelengths of different diseases were first extracted from the leaf spectral data using the RELIEF-F algorithm. A single band extracted from the most relevant wavelengths and the normalized wavelength difference from all possible combinations of the most relevant and irrelevant wavelengths were used to form the optimized spectral indices, which were then used to distinguish healthy and diseased winter wheat leaves. The classification accuracies of these indices on healthy leaves and leaves infected with powdery mildew, yellow rust, aphids were86.5%,85.2%,91.6%, and93.5%, respectively. The new spectral indices were also applied to non-imaging canopy data of winter wheat, and the classification of different diseases was satisfactory. Moreover, in leaf scale, the powdery mildew-index (PMI) had a good correlation with the disease index (DI), and therefore this index can be used to reversely identify powdery mildew severity. For canopy scale, the detection of severity of yellow rust through the yellow rust-index (YRI) showed high coefficient (0.86) between the estimated DI and actual DI values, suggesting that the new spectral indices can potentially improve disease detection and identification in precision agriculture applications.