Non-destructive Diagnosis Of Leaf Nitrogen Concentration By In-field Visible/near-infrdared Spectroscopy In Pear Orchars

Nitrogen is the basis of pear growth and development as well as yield formation.Real-time,non-destructive monitoring of pear nitrogen nutrition status and improving nitrogen utilization is not only the requirement of normal growth and development of pear trees,but also is the urgent need to improving the modern agricultural production.Visible/near infrared spectroscopy is considered to be a continuous and high resolution spectroscopy,which provides new means for plant nutrition diagnosis.Recent years,with the application of visible/near infrared spectroscopy in crop growth monitoring,its rapid,non-destructive and accurate advantages can replace conventional chemical analysis methods.Based on the field experiment,the quantitative and near-infrared spectra of pear leaves under different conditions,different growth stages,different nitrogen application rate,different geographical and cultivars were analyzed,and the quantitative correlation between leaf nitrogen concentration and leaf spectra was established.Nondestructive diagnostic model system for leaf nitrogen concentration was established by the Visible/near infrared spectra of pear leaves.And then based on different nitrogen treatments,the nutrient status of pear trees were quickly determined by using the established non-destructive diagnostic model in the fruit enlargement period.The nitrogen fertilizer was recommended by the nitrogen fertilizer formula as well as the results of the non-destructive diagnosis model.The results are as follows:1.The results showed that the visible/near infrared spectral curves of pear trees were smoothing with good stability by using the vegetation probe and the blade holder,and the signal-to-noise ratio was better than that of the 25° bare fiber method.The correlation coefficients between the reflectance spectrum measured by the vegetation probe and the black background/white background of the blade holder were analyzed.The results show that the correlation coefficient between the reflectance/transflectance and the leaf nitrogen concentration are very close.However,the correlation coefficient between the leaf reflectance and the leaf nitrogen concentration in the short-wave infrared region is slightly higher than that of the transflective spectrum.The signal-to-noise ratio by different measurement methods is:25° bare fiber<vegetation probe with blade holder of white background<vegetation probe with blade holder of black background.The relationship between leaf visible near infrared spectra and nitrogen concentration in different growth stages was discussed,and the leaf nitrogen concentration and its correlation with yield were also investigated at different growth stages.Results showed that the correlation between the leaf nitrogen concentration and the yield was found to be the highest at the 50 days after bloom.The leaf nitrogen concentration at the 50 day after bloom was found to be the highest during the growth season.The best diagnostic period is found to be 50 days after bloom.2.The spectra of 710 leaves were divided into 4 groups according to the total nitrogen concentration from low to high.The reflectance and transflectance of the leaves were obtained by averaging the spectra of the leaves with different nitrogen concentrations.We analyzed the visible/near infrared reflectance and the transflectance curves of pear trees with different nitrogen concentrations in different band regions.It was found that the spectral values of the leaves were significantly decreased with the increase of leaf nitrogen concentration at 350-700 nm.The spectra value in the near-infrared region of 750-1350 nm also decreases with increasing nitrogen concentration.The correlation between the leaf nitrogen concentration and the whole spectrum of reflectance and transflectance were analyzed and compared.It is found that reflectance and transflectance performed the same trend.The highest correlations were found at the 550 nm green peak and near the red edge of 720 nm.The accuracy of the model by the maximum correlation wavelength(705 nm)was low,who's R2 was only 0.2858.Furthermore,the non-destructive diagnosis model of leaf nitrogen content in pear leaves based on vegetation index was constructed.It was found that the difference spectal index between the leaf reflectance at 2170 nm and 2150 nm was the best,who's R2 and RMSE are 0.4586 and 2.19 g kg-1,respectively.3.The effect of the reflectance and the transflectance of the pear leaves on the accuracy of the model by four modeling methods was compared and analyzed.It is pointed that the reflectance can improve the model accuracy and its signal-to-noise ratio is higher than that of the transflectance.Among the four chemometrics modeling methods,the modeling accuracy of partial least squares regression(PLSR)and the back-propagation neural network(BPNN)are significantly higher than the principal component regression(PCR)and stepwise multiple linear regression(SMLR).The prediction accuracy of the PLSR is higher than that of the BPNN,and the BPNN is prone to overfit in the modeling of a single cultivar.On the basis of the original spectrum,spectral transformation and pretreatment are done to explore the effect of modeling and prediction on the accuracy.The results show that the original spectrum can improve the accuracy of modeling and prediction under normalized pretreatment.On the basis of this,a two-year single-species sample set was used to establish the partial least-squares regression model.The principal component of the model was 15,the determination coefficient(R2)of modeling and prediction were above 0.85.The root mean square error of calibration(RMSEC)was 1.3 g·kg-1 and the root mean square error of valication(RMSEV)was 1.5 g kg-1.The performance of the model can meet the demand of non-destructive determination the single species leaf nitrogen concentration of pear trees.4.The characteristics of visible/near infrared reflectance spectra of pear leaves in different years,different regions and different cultivars were compared and analyzed.The results showed that the visible/near infrared spectroscopy curves of different varieties of pear trees with the same nitrogen concentration were almost the same,but their numerical valueswere different in specific absorption/reflection peaks.The specific absorption reflectance peaks of visible and near infrared reflectance spectra of different regions and years was different.The difference of nitrogen concentration among leaves of different cultivars was significant.The correlation between the visible and near infrared spectra and the leaf nitrogen concentration of Yali,Kotobuki Shinsui and Cuiguan was found to be a similar trend.However,the correlation coefficients between the leaf nitrogen concentration and the original spectrum of Huangguan and Yuanhuang were found to be significant different.Based on the signal enhancement ability of Adaboost algorithm,it was inputted firstly into the BPNN and the support vector machine(SVM)algorithm.The non-destructive diagnostic model of mixed pear leaf samples with different regions,different varieties and different years was constructed.And the four classification modeling methods were compared and analyzed.The results show that Adaboost combined with BPNN has the best modeling accuracy and modeling stability.The R2 of modeling and prediction are both larger than 0.9,the mean square error(MSE)is less than 1.5g·kg-1,and the mean relative error(MRE)is less than 4%.The model performance can meet the demand of non-destructive determination of pear leaf nitrogen concentration by the different regions,different varieties,and different years.5.Based on the different nitrogen application,the nitrogen nutrition status of pear trees was quickly determined by using the established non-destructive diagnostic model in the fruit enlargement period,and the nitrogen fertilizer was recommended both by the nitrogen fertilizer formula and the non-destructive diagnosis of pear nitrogen nutrition.The results showed that the regulated nitrogen application could increase the leaf nitrogen.content under low nitrogen treatment and decrease the leaf nitrogen content under high nitrogen treatment.The effect of control treatments was remarkable.Under the low nitrogen level,the effect of topdressing nitrogen on the topdressing nitrogen was not obvious or decreased,but the nitrogen application rate was lower than that of the control.In conclusion,our results illustrated that it is a recommended remedy for good yield and quality to use this N management by the visible/near infrared spectroscopy especially in the condition of nitrogen deficiency.Accordingly,other kinds of fertilizers should be adjusted properly cooperating to the N management.6.Finally,we used Matlab R2012b as the development tool,M as the development language,combined with the existing spectral data processing methods and existing sample set to design and develop the non-destructive diagnostic system of pear leaf nitrogen concentration.The system is a software application which can be used to model the visible/near infrared reflectance spectra and the leaf nitrogen concentration of pear trees and to realize the non-destructive diagnosis of leaf nitrogen concentration.The non-destructive diagnostic system of pear leaf nitrogen concentration contains not only pretreatment methods and related modeling methods for dealing with the leaf spectrum of pear trees,but also a sample set which includes five widely cultivated varieties.The software can directly output the leaf nitrogen concentration of the five varieties by inputting the leaf visible/near infrared reflectance.And it can also establish a model independently according to the various methods of modeling methods.As a result,the accurate estimation of the leaf nitrogen concentration of pear trees can be achieved.In this study,non-destructively monitoring of leaf nitrogen content was realized by using visible/near infrared spectroscopy in pear orchards.On this basis,nitrogen fertilizer management was established based on visible/near infrared spectroscopy.