Grading Detection Technology Of Rice Blast Based On Hyperspectral Remote Sensing

In this study,the hyperspectral data and chlorophyll content of 180 tillering rice leaves infected by rice blast were measured by portable spectrophotometer and chlorophyll meter.45 samples were randomly selected as the training set and 45 samples were used as the test set to study the spectral characteristics of the original spectrum,derivative spectra and logarithmic spectra of different grades of rice blast.The relationship between the rice leaves infected with rice blast and the leaves of healthy leaves was discussed The method of grading of rice blast and the prediction model of chlorophyll content based on sensitive band were proposed by using neural network technology,which provided the basis for monitoring rice blast.The conclusions are as follows:(1)First order derivative spectrum,second derivative spectrum,logarithmic spectral transformation and other mathematical transformation to eliminate the background environment noise has a good effect.The results showed that the "blue edge position" and "red edge position" were significantly correlated with the disease severity of the rice,which could well reflect the health status of the rice leaves.(2)The sensitivity bands of the original spectrum were 473 nm,494nm,666 nm,674nm,700 nm,752nm respectively by the sensitivity analysis method,the spectral characteristics and the chlorophyll content correlation analysis,and the sensitive bands of the first derivative The sensitivity of the second order derivative spectrum is574 nm,665nm,701 nm,and the sensitivity of the logarithmic spectrum is 540 nm,570nm,670 nm,772nm,783 nm.(3)The sensitive band as input,rice blast level as output.A series of 135 training set samples and 45 test set samples were identified by system clustering,BP neural network and probabilistic neural network.The results show that the accuracy of logarithmic spectral grading is 97.8% in the training samples based on probabilistic neural network classification.The precision of logarithmic spectral grading is 75.5%.(4)Using the method of multiple stepwise regression,radial basis function neural network and spectral characteristic parameter method to construct chlorophyll content estimation model of rice leaf blight affected by rice blast.Among them,the mean square error between the predicted value and the measured value is 0.83,andthe average relative error is 7.5% by the regression model constructed by the multiple stepwise regression analysis.The root mean square error between the predicted value and the measured value is tested by the regression model constructed by the radial basis function neural network.The average relative error is 8.7%.The regression model constructed by the spectral characteristic parameters tests the root mean square error between the predicted value and the measured value of 1.27,and the average relative error is 10.2%.