Rapid And Real-time Estimation On Maize Blight Area By Using The Hyperspectral Image Technology

Maize is one of the main food crops in our country.In recent years,the phenomenon of maize reduction caused by leaf blight disease has became more serious.It is time consuming and laborious to rely on traditional method of field investigation and manual identification of maize leaf blight disease severity,and this method is qualitative and subjective.Hyperspectral imaging technology has the characteristics of real-time,fast,non-destructive and quantitative acquisition of target attributes.Thus,hyperspectral imaging technology has important research value for accurate monitoring of maize leaf blight disease severity.In this study,leaf blight diease area and hyperspectral imaging data was obtained through based on the investigation and collection of 82 maize leaf blight disease leaf samples,in order to explore the feasibility of using hyperspectral imaging technique for quantitative monitoring of maize leaf blight disease,the results are as follows:(1)The different degree of spectral reflectance in maize leaf blight disease had similar rules:spectral reflectance showed a slow upward trend in the 900?1350 nm,and obvious absorption valley in the range of 1350?1650 nm,and the differences maize leaf blight disease degree has the obvious different characteristics,suggested imaging spectra had sensitive response of maize leaf blight disease.The spectral reflectivity of maize leaf blight disease increased with the increase of disease areas are gradually increasing.And the variation of spectral reflectance curves also coincided with the physiological changes of maize leaf blight disease.(2)Compared with the original spectrum,1st,2nd and Noise pretreatment can obviously change the spectral curve features,the correlation coefficient of MSC and SNV pretreatment spectrum more than 0.7 at the wavelength range of 896-1110 nm,1259?1431,the highest correlation coefficient absolute value was 0.89,showed that spectral pretreatment can obviously improve the correlation of imaging spectra and maize leaf blight disease area.(3)it is proved that the spectral bands of 896?1110nm,1393?1571nm and 1650nm had a more important relationship with maize leaf blight disease.(4)Through building linear spectral model(SPA-MLR)of characteristic band,and the linear spectral model of full spectrum information(PLSR)and nonlinear spectral model(SVM)based on maize leaf blight disease,and compared the model performance showed that the SVM model spectrum as a whole has a better performance,thus proved there was a nonlinear relationship between maize leaf blight disease and hyper spectrum.However,spectral characteristics model of SNV pretreatment had the best performance(Rc2=0.954,RMSEc=0.036,RPDc=4.601;Rv2=0.912,RMSEv=0.043,RPDv=3.585),which can achieve the accurate estimation of maize leaf blight disease.