Fast Diagnosis Of Sclerotinia Infected Oilseed Rape(Brassica Napus L)Based On Hyperspectral Image

Precision agriculture is the core technology to realize the sustainable development of modern agriculture. How to obtain accurate farmland information is the key point of the development of the precision agriculture in the future. This study is mainly focused on the oilseed rape leaf (Brassica napus L). Based on hyperspectral imaging system, this study develops the early diagnosis model of oilseed rape leave infected Sclerotinia. It is meaningful for the precision production of oilseed rape. The main results were achieved as follows:(1) The spectral recognition models and imaging recognition models were developed for early diagnosis of sclerotinia sclerotiorum disease of oilseed rape leaf. Within visible spectrum range and near infrared spectrum range, set up partial least squares (PLS) and support vector machine (SVM) model based on full spectrum which was preprocessed by moving average. Effective wavelengths (EWs) were selected by different methods, such as weighted regression coefficient, PCA-loadings,x-loading weights and derivative-2. Based on different effective wavelengths, established PLS, linear discriminant analysis (LDA), back propagation neural networks (BPNN) and extreme learning machine (ELM) models and compared recognition ratios. Within visible spectrum range, the optimal recognition ratio was achieved by effective wavelength-extreme learning machine (EW-ELM) model. Within near infrared spectrum range, the optimal recognition ratio was achieved by effective wavelength-back propagation neural networks (EW-BPNN) model.(2) A discriminant model was established for sclerotinia detection based on effective vegetation indexes. Calculated the Pearson correlation coefficient between vegetation indexs and oilseed rape leaf status by using SPSS Statistics. Based on a single vegetation index, established the LDA model and chose the most effective6vegetation indexes. Based on effective vegetation indexes, established PLS, SVM, BPNN and ELM models. It had the best performance.(3) The physiological information of Catalase (CAT), peroxidase (POD) and Superoxide Dismutase (SOD) of oilseed rape leaves infected sclerotinia was calculated by spectral technology. According to species and enzyme activity different, selected different sample spectrums to calculating set and predicting set. For a single enzyme, based on the raw spectrum and7kinds of spectra preprocessed method (Moving average-7, SNV,1-Der,2-Der, De-trending, MSC and Baseline), established PLS model and SVM model. Selected effecvtive wavelength by xmodel based on full-spectrum (MSC) had a best prediction performance. Rp=0.6831, RMSEP=2.-Loading weights, and establish PLS and SVM model.①For CAT enzyme activity, the results indicated that PLS model based on full-spectrum (SNV) had a better prediction performance. The correlation coefficient of prediction (Rp) is0.7938, the root mean squares error of prediction (RMSEP) is0.9336. Based on effective wavelengths, PLS model had an optimal performance. Rp=0.7622, RMSEP=0.9659.②For POD enzyme activity, the results indicated that PLS8293. For effective wavelength, PLS model had an optimal performance. Rp=0.5960, RMSEP=3.0930.③For SOD enzyme activity, the results indicated that PLS model based on full-spectrum (SNV) had a best prediction performance. Rp=0.7834, RMSEP=198.7432. For effective wavelength, PLS model had an optimal performance.Rp=0.7999, RMSEP=185.1372.