| In this study,we used ASD Field Spec 3 portable spectrometer collected the spectrum of disease tobacco leaves(Zhongyan 100)under natural conditions three times in vigorous growth period of tobacco,and determined the physiological and biochemical indexes of disease leaves under indoor conditions at the same time.The characteristics of original and differential spectral,correlation of hyperspectral characteristic variables and the physiological and biochemical indexes of disease tobacco leaves weas analysed through quantitative and qualitative method.The tobacco mosaic virus damage severity of discriminant analysis models were finally established by screening and comparison different chemometric modeling pre-processing methods,which in order to provide theoretical support for the realization of large area synchronization and non-destructive monitoring of tobacco mosaic disease by using remote sensing technology in the future.The main conclusions of this study are showed as follows:1、With the aggravation of the disease degree,the spectral reflectance of tobacco leaves infected by tobacco mosaic disease was gradually increased in the whole wave band compared to healthy tobacco leaves,the difference is more obvious in Near-infrared(760~1300 nm)and the short-wave infrared(1300~2500 nm)region especially.The first derivative spectrum of disease tobacco leaves was of greatest change in the red edge area of visible light(680~760 nm),and the position of the red edge was gradually shifted to short wave direction,then the "blue shift" phenomenon occurred.2 、 The physiological and biochemical indexes of tobacco leaves were significantly different in healthy and different disease degree of tobacco leaves,among which the content of pigment,water,total sugar,reducing sugar and total nitrogen decreased gradually with the increasing of disease degree,while the nicotine content increased gradually.By means of the correlation analysis between the physiological and biochemical indexes and the original,first derivative spectrum and the high spectral characteristic variables,sensitive wave bands and hyperspectral characteristic variables of the physiological and biochemical indexes were screened out.The physiological and biochemical indexes stepwise regression models of diseased leaves were built,and the quantitative estimation of physiological and biochemical indexes of diseased leaves was realized.3、Visible and Near-infrared hyperspectral data in the range of 400~2400 nm were extracted with the method of chemical analysis,and the effects of different pretreatment methods on the accuracy of the models were systematically compared.It found out that the best Savitzky-Golay convolution smoothing points was five by comprehensive comparison,and the accuracy of model with differential treatment was improved in a large extent compared to raw spectral,which the first derivative treatment was superior to the second derivative treatment.First derivative(1-Der)combined with orthogonal signal correction(OSC)was the best pretreatment method on modelling,and the determination coefficients(R2)of calibration set and the validation set was 0.989 and 0.965 respectively,the root mean square error of calibration(RMSEC)was 0.082,the root mean square error of prediction(RMSEP)was 0.098.The load-factor method was used to extract the characteristic bands of the first eight principal components at the same time,and the least squares support vector machine(LS-SVM)and partial least squares discriminant analysis(PLS-DA)model were established respectively to analyze the degree of tobacco mosaic disease.The results showed that the overall recognition rate of calibration set and validation set was 96.50% and 90% in the LS-SVM model based on the feature band respectively;the overall recognition rate of calibration set and validation set was 100% in the PLS-DA model based on the feature band respectively,which performed the best and realized the accurate discrimination of tobacco leaves with different disease degree. |
PDF Full Text Request