Fast Diagnosis Of Nitrogen Status In Pear Leaf Using Spectroscopy Technique

Nitrogen not only is a very important nutrient for crop growth, yield formation and quality benefits, but also play a crucial role for timely determination of nitrogen status to ensure the crop growth. Recently, spectra techniques in plant nutrition diagnosis is becoming a very important research focus for real-time determination of leaf nitrogen content and necessary trend to substitute the conventional chemical analysis methods.In this paper, we firstly analyzed the spectral characteristics of fresh and crushed pear leaves, and then researched the impact factors from the collected spectral information, sample selection, sample status, and model parameters, respectively. The results showed that:(1) The reflectance spectra of pear fresh leaf acquisited by plant probe and the whiteboard has the advantage of low noise and high smooth, which is beneficial to increase the accuracy of prediction model. Spectral reflectance collected by the center of a leaf because it can better reflect the reflectance on the entire leaf. It’s important to note that the dust on the leaf surface have a significant impact on spectral reflectance and vegetation indices, and therefore it must be removed when acquisiting the spectrum.(2) The maximum sensitive wavelength of pear leaf nitrogen content is slightly different with the different growth stages. The maximum sensitive wavelength in May, July and August were715,710and718nm, respectively. The linear correlation coefficients established by the corresponding nitrogen content and wavelength reflectance were-0.605,-0.6186and-0.7936, respectively. The whole band (350-2500nm) and the original spectrum is most suitable for establishing pear leaves partial least squares prediction model.The principal components of the prediction model were11,13and12in these three periods when the accuracy reached the highest point, and the cross-validation of the model coefficient (R2cv) were0.7381,0.7584and0.8977, with Root mean square error of cross-validation (RMSECV) is1.815,2.005and0.9920g kg-1in the three periods, respectively. The average errors of the unknown samples were3.78%,6.05%and3.30%predicted by the three models, respectively. On the basis of the whole band and the original spectral form, the accuracy of nitrogen predict model was highest with the mothod of partial least squares after the whole growth period of the spectrum by the moving average smooth (segment size=3) after pretreatment. At this time, the principal component was10, and cross-validation coefficient and root mean square error were0.9520and1.417g kg-1, respectively.In addition, the predicted average error of unknown samples was4.48%, less than5%, and therefore this model can well predict the pear leaf nitrogen content.(3)Through the analysis of the nitrogen content in crushed leaves and mid-infrared photoacoustic spectroscopy, we established the prediction models in the experiment stations of Yantai, Tai’an, Yingkou, Changli and the total of the four stations. The results showed that the prediction error of the models ranged from1.26-2.18g·kg-1, and the predicted correlation coefficients ranged from0.644-0.806. Therefore, there is a high practicality for prediction with this method. The order of the accuracy in the five prediction models is Yantai> Tai’an> Yingkou> Total> Changli. The prediction error and correlation coefficient in the Station of Yantai was1.92g kg-1and0.806, respectively.Therefore, the model of rapid determination of freshed pear leaf nitrogen content can be established through simutaniusly select the method of visible and near-infrared reflectance and chemometric method, while the rapid determination of crushed leaf nitrogen content can be modeled through simutaniusly select mid-infrared photoacoustic spectroscopy and chemometrics method. All of the above two prediction models can meet the need for fast and convenient detection of pear leaf nitrogen content.