Research On Diagnosis Of Rice Sheath Blight Disease Based On Hyperspectrum

Rice disease monitoring and diagnosis is of great significance to ensure national food security.Detection which was based on molecular biology can realize precisely analysis of disease,but it costs high,has low efficiency,needed technically operation,so that it is difficult to apply in agricultural production.The application of spectral technology in disease can provide high-throughput,rapid detection of rice disease in recent years.In this research,spectral response characteristics of japonica rice in northeast China with rice sheath blight diseases,are used for analysis of disease identification,disease hierarchies,disease index estimating etc,the main content are as follows:(1)This study identified whether the single plant of rice was infected with sheath blight or not without images on leaf scale.The results show that three spectral feature extraction methods can effectively extract the characteristic spectral information of rice sheath blight disease.The Gram-schmidt orthogonalization method was beneficial to the improvement of the recognition accuracy of rice,with the recognition accuracy of 90%.At the same time,from the perspective of data analysis,the sensitive feature wavelength was located at 550nm,675nm and 730nm.(2)This paper also includes identification and estimation of disease severity of rice sheath blight on leaf scale,The response to sheath blight were analyzed,and the differences of between infected and healthy conditions were compared with the result of sensitive wavebands were 650-700nm and 720-820nm.Disease level recognition on leaf scale is based on the original spectrum,the original spectrum of first order differential value,relevant vegetation index,the original spectrum continuum removal mode,and basis vectors found by Gram-Schmidt orthogonalization,the results show that base vector can effectively extract the spectral information,on the qualitative analysis is also superior performance.(3)This paper has found that the characteristic of the original rice hyperspectral data without images on canopy scale from the whole data trend and some important wavelengths,which were the core of rice sheath blight disease identification,the canopy scale spectral data were sensitive to rice sheath blight at 583-635nm and 655-741nm.The disease index of rice disease on canopy scale characteristics were used as independent variables to establish the disease index estimation model.The results show that the unitary linear model constructed with the basis vector of the original spectrum as the feature as the input variable is the best model for disease severity inversion.Compared with other features,the results showed basis vector was a better feature selection method for rice canopy hyperspectral data as well,and the estimated accuracy R~2is 0.675,which has realized the detection of canopy blight at near-surface level.The results can improve the accuracy of non-destructive detection of rice sheath blight disease,lay a theoretical foundation for the development of rapid detection instruments of rice disease.