Detection Mechanism And Methodology Of Brassicanapus Disease Using Spectroscopy And Spectral Imaging Technologies

In this study, hyperspectral imaging, chlorophyll fluorescence imaging, laser induced breakdown spectroscopy, mid-infrared spectroscopy and laser confocal micro-Raman imaging were used to detect oilseed rape Sclerotinia stem rot. The main conclusions were as follows:(1) Detection of Sclerotinia stem rot of leaves and stems of different sample sets by hyperspectral imaging was studied. The average spectra, pixel-wise spectra and corresponding vegetation indices combined with discriminant models were used. The results showed that the spectral preprocessing methods did not have significant influence on the discriminant results, and the selection of discriminant models were more important for Sclerotinia stem rot detection. The simulation of early inoculation of Sclerotinia stem rot using the region near the infected region showed good results, indicating the feasibility of using hyperspectral imaging for early detection of Sclerotinia stem rot. The results showed that optimal wavelengths selected by 2nd spectra and PCA loadings had good repeatability and reproducibility among different sample sets. In all datasets of different sample sets, ELM、RBFNN、SVM and RF models showed good discriminant results. Determination of pigments and soluble protein content in oilseed rape leaves using hyperspectral imaging showed that both optimal wavelength selection methods and regression models showed different performances in different sample sets, and the results showed that PLSR and ELM obtained good prediction performances.(2) Detection of Sclerotinia stem rot of leaves and stems of different sample sets by chlorophyll fluorescence imaging was studied. The distribution of chlorophyll fluorescence parameters within a leaf and a stem was investigated. The results showed that the chlorophyll fluorescence parameters of healthy regions were different from the chlorophyll fluorescence parameters of infected regions, and the chlorophyll fluorescence parameters of the regions near the infected regions showed differences from the other 2 regions, and the stem had smaller differences. Optimal chlorophyll fluorescence parameters were selected by correlation analysis, ANOVA and LDA, the results of discriminant models using optimal chlorophyll fluorescence parameters indicated the feasibility of detecting Sclerotinia stem rot using chlorophyll fluorescence imaging, the results showed that SVM、KNN、NBC and RF models obtained good performances.(3) Detection of Sclerotinia stem rot of leaves of different sample sets by laser induced breakdown spectroscopy (LIBS) was studied. The LIBS spectra of fresh leaves showed obvious noises, and WT denoising could significantly reduce the noise level. Robust baseline estimation showed the best performance for baseline correction of LIBS spectra. The optimal LIBS wavelengths were selected and analyzed by 2nd spectra, PCA loadings and Bw, The selected optimal LIBS wavelengths matched with the peaks of the origin spectra. The results of discriminant models using full spectra and selected spectral peaks indicated the feasibility of using LIBS to detect Sclerotinia stem rot. The overall results showed that the spectral peaks could be used for analysis instead of full spectra, and ELM, RBFNN and RF models obtained good performances.(4) Detection of Sclerotinia stem rot of leaves of different sample sets by mid-infrared spectroscopy (MIR) was studied. WT denoising was used to remove the noises in MIR spectra. Six different methods were used to select optimal wavenumbers, the optimal wavenumbers selected by 2nd spectra and PCA loadings had good repeatability and reproducibility among different sample sets. The results of discriminant models using full spectra and the optimal wavenumbers showed that PLS-DA, RBFNN, ELM, SVM and RF models obtained good performances.(5) Detection of Sclerotinia stem rot of leaves by laser confocal micro-Raman imaging was studied. The Raman spectra of sampling points in healthy, regions near infected regions and infected regions of infected leaves were studied. The background information of Raman spectra were removed by using baselineWavelet algorithm. The results of discriminant models using full spectra and the spectral peaks indicated the feasibility of using Raman spectroscopy to detect Sclerotinia stem rot. The overall results showed that the spectral peaks could be used for analysis instead of full spectra, ELM, RBFNN, RF, SIMCA, SVM and KNN models obtained good performances. The slopes of Raman spectra were studied for Sclerotinia stem rot detection, and the slopes of Raman spectra of healthy, regions near infected regions and infected regions showed significant differences in ranges and distributions, which could be used for Sclerotinia stem rot detection.