Early Detection Of Rice Leaf Blast From Leaf-level Imaging Spectroscopy Data

Rice blast,known as rice cancer,causes huge economic losses to rice production every year.China has been advocating the concept of green prevention and control of crop diseases for many years,and remote sensing has became an indispensable scientific means for dynamic monitoring and precision management of crops with its objective,real-time and non-destructive advantages.However,there are still some problems in the monitoring of rice blast at present:relavant studies lack early monitoring of rice blast.;using non-imaging hyperspectral data to monitor rice blast lacks spatial information,which is not conducive to the spatial identification of rice blast;by selecting sensitive bands,the method of constructing empirical model encounters a series of problems that are insufficient robustness due to changes in crop varieties and cultivation environment.The method of inversion of biochemical parameters by radiation transfer model for rice blast monitoring has high stability and portability,but the traditional inversion method of radiation transfer model is inefficient and not practical.In this study,firstly,we analyze the temporal and spatial of the spectral changes of infected pixels,and spectral differences between large-volume healthy and infected samples.The results show that on the first day of the occurrence of rice blast disease,the spectra of the infected samples and the healthy samples shows a significant difference,and using the wavelet coefficient spectrum can capture more details than the reflectance spectrum.This study based on the PROCWT model to optimize the LUT,and explore various strategies to improve the accuracy of the LUT inversion,such as changing the scale of continuous wavelet transform,selecting multiple solutions and adding Gaussion noise to spectrum.According to this research,the parameters with highly linear correlation in the look-up table take the "joint uniform distribution in the confidence interval",which is the best for inversin of infected rice leaf.After using optimized lookup table inversion strategies,in the inversion of rice leaf biochemical parameters,the RMSE of Canth and Cchl is 2.85μg/cm2 and 0.33μg/cm2 respectively,and the single spectral inversion efficiency is 0.14 seconds.Finally,the spatial distribution of biochemical parameters of rice leaves were retrieved using the optimized lookup table inversion strategy.We found the spatial distribution of pigment content in infected rice leaves could serve as a guide for identifying the distribution of disease spots.Thus,we took the parameters inverted by optimized lookup table as input features of SVM to identify and classify the disease spots of healthy and susceptible rice leaves.The results show that on the first day of the occurrence of rice blast disease,rice leaf blast spot could be accurately identified from infected rice leaves.Based on PROCWT,this study summarizes a highly efficient and accurate lookup table inversion strategy that is suitable for infected rice leaves to invert biochemical parameters,and also identifies and classifies the leaf blast spot based on SVM and rice leaf biochemical parameters.The monitoring methods of rice leaf blast were systematically explored at single leaf scale.This monitoring of rice leaf blast has important implications for the extension of the canopy scale and for the study of other crop diseases.