Identification And Warning Of Rice Leaf Blast Based On Analysis Of Chlorophyll Fluorescence Spectrum

In recent years there is a growing interest in monitoring of vegetation byspectroscopic detection of electromagnetic radiation. It’s a kind of nondestructive andrapid method in the study of environment. Laser induced chlorophyll fluorescence（LICF） spectroscopy of terrestrial vegetation is an important aspect of activedetection which provides a specific way for assessing the status of vegetation inenvironment. In order to achieve the detection of rice leaf blast accurately and rapidly,analysis of chlorophyll fluorescence spectrum was taken as the main means,identification and early warning models of rice leaf blast were strived to be buildbased on the analysis of environmental information, rice physiological informationand biochemical information.The rice leaf blast was divided into different grades and the chlorophyllfluorescence spectra were applied to identify the disease. According to the standardevaluation system for rice formulated by International Rice Research Institute (IRRI)and the characteristics of disease, the rice leaf blast was divided into three grades andwas determined the levels of the measured leaves on the basis of the area ratio is equalto the pixel ratio. First of all, chlorophyll fluorescence spectrum and diseasecharacteristics were obtained at the same time using a compact fiber-opticfluorosensor with a solid-state laser at473nm as exciting source and an integratedfiber-optic spectrometer at ambient temperature. After the basic treatment with thespectrometer’s own AVANTES Software7.2, chlorophyll fluorescence spectrum wererespectively dealt with Savitzky–Golay(SG) smooth, SG smooth+First DerivativeTransform(FDT), SG smooth+Fast Fourier Transform(FFT) and WaveletDe–noising(WD) for spectral preprocessing. Then Principal Component Analysis(PCA) and Gaussian Function Fitting(GFF) on chlorophyll fluorescencespectrum under different pretreatment were applied for dimension reduction.Furthermore, Discriminant Analysis(DA), Multiple Logistic RegressionAnalysis(MLRA), Back Propagation Neural Network(BPNN), Support VectorClassification(SVC) machine based on the four classical kernel function were used toidentify and compare the rice leaf blast at the three levels. The results show that theidentification accuracy of GFF-SVC model based on PLOY kernel function andPCA-SVC model based on PLOY and RBF kernel functions are more efficient thanothers as it’s up to95.0%. The results also show that the combination of differentbands can take different influence on the results, so we can choose different bands fordifferent purpose in the identification of rice leaf blast.In order to adopt more methods to identify rice leaf blast and improve theidentification accuracy of rice leaf blast, four classic kernel functions were combinedin different ways to create new kernel functions which take the components ofspectrum with SG-FDT pretreatment as the input. Then comparative analysis wasmade between the new kernel functions and the classic kernel functions. The resultsshow that the linear combination kernel functions have a better performance than theclassic kernel functions. Especially the kernel function of PLOY+RBF is in the bestperformance to the all as its identification accuracy is up to100%for level0.Although the performance of RBF+Sigmoid kernel function is the same asPLOY+RBF kernel function for level0, but the identification efficiency is lower thanthe PLOY+RBF kernel function for level1and level2. While the performance ofSigmoid kernel function is the lowest for the identification of rice leaf blast, but itscombination with PLOY and RBF kernel function respectively can effectivelyimprove the identification efficiency of rice leaf blast.To guide agricultural production, early warning of rice blast is taken as a mainmeans. In the research, environmental information, rice physiological information andchlorophyll fluorescence spectrum information were applied to the rice leaf blast earlywarning. First of all, a weather station was placed within the rice field from June1to September15,2012and2013for monitoring the environmental information and itwas set to transmit information once every1hours. Qualitative analysis was carriedout on the environment temperature and humidity, the daily highest temperature andhumidity, daily minimum temperature and humidity and daily average temperatureand humidity were chosen for the building of rice leaf blast early warning model. Theearly warning model was established with the information collected in2012and theinformation collected in2013was used to validate the prediction ability of model.The results show that the environmental temperature and humidity can be used as awarning sign index. Then leaf physiological information and chlorophyll fluorescencespectrum information were used for rice leaf blast early warning. Physiologicalinformation mainly includes the light use efficiency (LUE), water use efficiency(WUE) and SPAD values. The relationships between LUE, WUE and F685, F732,temperature of leaf (Tl) were analyzed respectively and linear models were set up topredict LUE and WUE. PLSR, PCA-BP, PCA-SVR(RBF), PCA-SVR(PLOY) modelswere established for the prediction of LUE、WUE and SPAD values which take thecomponents of spectrum with SG-FDT pretreatment as the input. Comparing theprecision of different models, the best model was chosen on the basis of the predictionaccuracy to the prediction set.Biochemical information early warning models of rice leaf blast based on theanalysis of chlorophyll fluorescence spectrum were established. The first antioxidantenzyme named superoxide dismutase(SOD) which plays a role in the process of activeoxygen removal and malondialdehyde(MDA), free proline(Pro) whose contentchanged as the microorganism infection were selected as the research objects for riceleaf blast early warning. Spores were vaccinated in rice leaves in the laboratory.Chlorophyll fluorescence spectrum and SOD, MDA, Pro were measured at0h,12h,24h,36h,48h and60h after the inoculation. In the research, early changes of SOD,MDA, Pro were analyzed and the PLSR, PCA-BP, PCA-SVM(RBF) andPCA-SVM(PLOY) models based on the chlorophyll fluorescence spectrum to predictSOD, MDA, Pro were established. The relationship between the predicted values and measured values was analyzed based on the prediction set. The results show that SOD,MDA, Pro could be used as the basis of rice leaf blast early warning. PCA-BP modelhas a good prediction effect for SOD as the PCA-SVM(PLOY) model is better forMDA and Pro prediction than others.Using the above models and analysis, the identification and early warning systemframework of rice leaf blast was built, mainly including the structure design, functionmodule design and workflow design. This framework can provide important supportfor the subsequently establishment of identification and early warning system toeventually meet the users’ query and identification or early warning for the rice leafblast.