Application of Raman spectroscopy for early detection of rape sclerotinia on rapeseed leaves
Zhao Yanru,Yu Keqiang,Li Xiaoli,He Yong
DOI: https://doi.org/10.11975/j.issn.1002-6819.2017.01.028
2017-01-01
Abstract:Raman spectroscopy technique has been widely used in detecting the physiological information of plants.Due to its unique advantages of simple pre-treating,rapid response,high sensitivity and in-situ,nondestructive detecting,it can be performed to acquire biological information.As an important oil crop,oilseed rape (Brassica napus L.) is widely cultivated all around the world.High energy and protein livestock feed are mainly made from its seeds.It is also partly used as potential raw material in synthesizing biodiesel.However,sclerotinia rot of colza,which is generally caused by fungal pathogens sclerotinia sclerotiorum,seriously limited the development of rapeseed industry.Therefore,early detection of the sclerotinia sclerotiorum infection on rape leaves will helpful to discriminate,prevent and cure the sclerotinia rot of colza on rape plants.In general,plant pathogen spread to the tissue around the invasion point,the different distance between the invasion points on the scab represents different severity of the disease on the host plants.In this study,a total of 90 oilseed rape leaves were collected for this experiment.90 Raman spectral curves of healthy,mid-infected and severe infected oilseed rape leaves were acquired by confocal micro-Raman spectroscopy in the region of 800-2 000 cm1.Baseline algorithms were employed to process baseline correction.Wavelet transform based on the time-frequency domain can undertake multi-scale decomposition of Raman spectra,therefore,fluorescent background would be removed by reconstructing the signal without low frequency signal.Herein,wavelet transform was used to remove fluorescence background from the original spectral information.Significant differences at 0.05 level among the three kinds of samples at three characteristic peaks (1 006,1 156 and 1 522 cm1) were analyzed by one-way analysis of variance (ANOVA) method.Then,principal component analysis (PCA),which can compress the representation of a collection of vectors,was adopted to process cluster analysis.93% information of the original data was represented by the first two principal components (PC-1 and PC-2).Then,variables of principal components (PC-1 and PC-2) and characteristic peaks (1 006,1 156 and 1 522 cm-1) were employed to process cluster analysis.Finally,least squares support vector machine (LS-SVM) model,which is highly adaptive,was established based on the data of principal components and the characteristic parameters to discriminate the infected severity.The results revealed that LS-SVM models based on the PC-1,1 156 and 1 522 cm-1 provided a discriminative accuracy of 100%,LS-SVM discriminative models based on PC-2 and 1 006 cm-1 with accuracy of 66.7% and 70.0%,respectively.The results proved that Raman spectra analysis on the disease spot is meaningful for the early detection of disease.Meanwhile,this research provided a theoretical reference for portable Raman instrument designing in disease infection's early detection.What is more,different varieties of oilseed rape plants infected by different diseases would be detected if Raman spectroscopy technique is applied.Raman spectra coupled with molecular analysis provide a promising way in detecting specific pathogen of plants at early infection stage in future study.