Early Prediction Of Potato Late Blight Using Hyperspectral Technique

As a destructive disease,potato late blight has threatened the sustainable and healthy development of potato industry.Taking potatoes under the stress of late blight as research subject and hyperspectral imaging technique as a method,this study develops the prediction models and variation of several physiological indexes in potato leaves and achieve instant and fast detection for physiological information in potato leaves with late blight.Meanwhile,the recognition models were built for early diagnosis of potato seeds with late blight in order to control disease at its source.It is meaningful for the prevention of potato late blight.The main research results were achieved as follows:(1)The spectral prediction models were established for the detection of SPAD value in potato leaves with late blight.Different preprocessing methods including moving average smoothing,multiplicative scatter correction(MSC),spectroscopic transformation(ST),normalize and derivation were compared in partial least square(PLS)and least square-support vector machine(LS-SVM)models.The Normalize-PLS model achieved the best performance.In order to simplify models and decrease inputs,successive projections algorithm(SPA)and x-loading weights(X-LW)were used to extract effective wavelengths and discuss for optimal model.The results showed that LS-SVM model based on effective wavelengths selected by SPA performed best.Furthermore,the vegetation indices which were good correlated to SPAD value were studied for prediction models.The quadratic polynomial model obtained the precise accuracy.(2)Based on hyperspectral technique,the prediction models were established for the determination of catalase(CAT),peroxidase(POD)in potato leaves infected late blight.For each enzyme,raw spectrum and other 7 kinds of different preprocessing methods were compared in PLS and LS-SVM models.Then effective wavelengths were extracted by SPA,X-LW and built PLS and LS-SVM models.(1)For CAT activity,the results indicated that PLS model based on full spectrum after median filter smoothing had the optimal performance Base on effective wavelengths,SPA-LS-SVM model achieved a better performance.(2)For POD activity,the results showed that LS-SVM model based on full spectrum after median filter smoothing preprocessing performed better.With effective wavelengths,X-LW-PLS model had the optimal performance.(3)The kinetic models of CAT and POD activity varying with accumulated days after infection in potato leaves was studied under the stress of late blight and achieved good prediction.(1)For CAT activity,the kinetic model had the form of CAT=8.998-.2906t+.1292t~2-.0173t~3and achieved good relativity.(2)For POD activity,the model wasPOD=36.036-24.639t+19.165t~2-.2859t~3with the high accuracy.The result indicated that both CAT and POD activity changes were not monotone increasing or decreasing,it is necessary that combing other external symptoms of potato late blight and enzymes activity to predict the degrees of disease.And CAT can be used to predict the potential risk of potato late blight.(4)The recognition models were developed for detecting healthy and diseased potato seeds under the stress of late blight.Normalizing was found to be the best method for spectral data preprocessing.Three feature extraction algorithms of SPA?PCA and X-LW were used to select effective wavelengths.Based on full spectrum,F-DA model obtained the optimal recognition ratio of 95.08%in prediction set.With effective wavelengths,SPA+F-DA model was performed best and the whole recognition rate in prediction set was 90.13%.