Inversion Of Winter Wheat Powdery Mildew Based On Hyperspectral Remote Sensing

In order to achieve early detection and precision prevention of winter wheat powdery mildew, it is necessary to clearly understand victimization levels and spectral characteristics at different levels of wheat powdery mildew stress. Experiments included field and greenhouse trials were separately conducted at Science and Technology Demonstration Park of Henan Agricultural University and National Engineering Research Center for Wheat during years of 2009-2010,2010-201 land 2014-2015. The tested materials were different winter wheat cultivars of powdery mildew resistance, Yanzhan 4110 and Aikang 58 etc. At different experimental conditions of bacteria infection levels, we studied spectral reflectance characteristics of the infected winter wheat at leaf and canopy scale using the monitoring method of vertical angle and multi-angle. We analyzed the correlation between spectral reflectance with disease severity or the disease index and the leaf nitrogen accumulation(LNA), and established the optimal spectral parameters, band combinations and the best monitoring model of DI and LNA under powdery mildew stress. The main results of this study as follows:(1) At the condition of leaf scale and vertical angle monitoring, we realized that the harm caused by powdery mildew on chlorophyll was greater than the internal structure, and the degree of response of the visible bands on wheat leaf disease severity was much more obvious than near infrared bands. The disease inversion model, constructed by the method of factor analysis and BP neural network (FA-BPNN), had high precision (R2val> 0.80)and little error (3.12%-12.01%). This model have better applicability for inversion situation of wheat leaf powdery mildew at different periods. The disease inversion model, constructed by the method of partial least squares regression (PLSR), had the best inversion effect of the early-middle filling stage successively than the whole filling and the middle-late filling. The coefficient of determination (R2val) and relative error (REval) of the three combination periods were 0.875,0.818,0.787 and 2.97%, 7.34%,12.72%, respectively. Therefore, The best monitoring period using PLSR to detect powdery mildew in wheat leaves was the early-middle filling stage.(2) At the condition of canopy scale and vertical angle monitoring, we pinpointed that modified Disease Index (mDI) was the suitable expression index indicated the incidence of wheat powdery mildew, it can reduce the impact of leaf area differences to spectral information. The spectral bands which sensitive to the infected wheat canopy were 550-710 nm. As for the correlation with mDI, the R2 of green bands combinations (R2>0.75) choosed by the method of two-band ratio index and normalized vegetation index were superior to conventional spectral parameters,in which double green simple ratio (DGSR (584,550)) and double green normalized difference (DGND (584,550)) have the same coefficient of determination and R2 were 0.845. Therefor, these two parameters had good effects in monitoring the powdery mildew occurs status during mid-later growing season of wheat.(3) At the condition of canopy scale and multi-angle monitoring, all bands of the main plane and vertical plane exhibited that reflectance of backward observations was greater than forward, and with angle from back to front, the main plane was more obvious than vertical plane. At the main plane, the four bands, included blue, green, red and near infrared bands, showed the shap of typical irregular "bowl" and with the DI increase, the shape of "Bowl" was more obvious. At the vertical plane, there was not only the shap of irregular "bowl", also exist less obvious "hummocky", even exist "hump-shaped". The positively correlated bands with DI of all the backward and forward observation angles of the main plane and the vertical plane were 582-704, 607-700,500-515,560-715 and 565-710 nm successively. The best viewing angle at different observation angles within 400-900 nm by the method of two-band ratio and normalized band combinations is -30°, the combinations was consist of 894 and 803 nm. The double near-infrared simple ratio (DNiSR(894,803)) and double near-infrared normalized difference (DNiSR(894,803)) had the same coefficient of determination, R2 were 0.886. These models can be used as the best monitoring angle and prediction model of powdery mildew severity.(4) At the condition of canopy scale and multi-angle monitoring, reflectance curves of LNA at blue, green and red bands all showed up backward observations was greater than forward, where the maximum and minimum reflectance angle was -60° nd 20°, respectively. The sensitive bands of LNA at different observation angles (r> 0.75) were 440-505 nm at -50°, 464-511 nm at -40° and 606-691 nm at -30°, respectively. The best parameters and angle to inversion LNA at backward, forward and vertical monitoring were Rmin (640-680nm) and -30°, PSRI and 0° and ND (SDr,SDy) and 50°, respectively. The best viewing angle at different observation angles within 400-900 nm by the method of two-band ratio and normalized band combinations is-50°, the combinations was consist of 437 and 497 nm. The coefficient of determination of double blue simple ratio (DBSR(437,497)) and double blue normalized difference (DBND (437,497)) were 0.862 and 0.859, respectively. This model with high accuracy can be seen as the best monitor angle and prediction model of wheat LNA under multi-angle obserations, thus it can achieve early identification of powdery mildew disease.