Study On Oilseed Rape Nutrition And Rapeseed Quality Analysis Based On Spectroscopy Technology

Precision agriculture has become a focus of concern for developed countries being confronted with21st century, utilizing reasonably agricultural resource, improving quantity and quality of agriculturalproducts, reducing producing cost, and improving environment and agricultural sustainabledevelopment. The essential of precision agriculture is precisely to manage complex agricultural systembased on information and knowledge. How to acquire field information rapidly and precisely hasbecome a difficult issue of precision agriculture.Considering the problem and deficiency of current technologies in the information acquisitionsystem and the actual situation of our country, oilseed rape growing and varieties and years of rapeseedinformation rapid acquisition were studied in this thesis. The main research work and achievement wereas follows:1. Quadratic regression orthogonal design was applied in the field experience according to thepre-experience in our laboratory. The equipments were determined by the feasible studies. A newmethod based on spectroscopy technology for the rapid detection of oilseed rape nutrition and thediscrimination of different varieties and years of rapeseed was investigated. Compared with traditionalmethods, it is non-destructive, rapid, non-contaminative, et al..2. The reflectance spectra of oilseed rape were explained, and also, linear and nonlinear models fornitrogen (N), phosphorus (P) and potassium (K) of oilseed rape were built by using differentchemometric methods. Least square support vector machine (LS-SVM) model combined with partialleast square analysis (PLS) showed the best result. The correlation coefficients for prediction ofLS-SVM model of N, P and K reached 0.9180, 0.8063 and 0.7043.3. The relations between reflectance of oilseed rape leaf and canopy spectra and chlorophyll ofoilseed rape were studied. With the analysis of correlation and regression coefficients betweenreflectance spectra and SPAD value of oilseed rape, the feature wavebands were found. PLS Models forSPAD value prediction were built by the full waveband, feature wavebands and feature wavelengths.The correlation coefficients for prediction of three models were 0.9407, 0.9299 and 0.7905. Thedifferent models for SPAD prediction were compared. The results showed that PLS-LS-SVM model was better than PLS and PLS-BPNN models. In addition, the PLS-LS-SVM model for SPAD valueprediction based on canopy reflectance spectra was also built, and the correlation coefficient reached0.8728.4. The discrimination of different varieties and years of rapeseed were studied by near infraredreflectance spectroscopy technology. The recognition correction was investigated by using differentpre-processing methods and calibration models. The results indicated that the model withSavitzky-Golay smoothing with five smoothing points combined with standard normal variatetransformation (SNV) obtained the best recognition ability. The recognition correction for prediction ofWT-BPNN and PLSDA-BPNN models for discrimination of rapeseed varieties reached 100%, and therecognition correction of PCA-BPNN model was 96.7%. The prediction precision of WT-BPNN modelfor discrimination of different years of Zheshuang 72 rapeseed reached 100%.