Estimation Of Nitrogen Content In Apple Leaves Based On Hyperspectral Imaging Technology

The traditional method for detecting the nitrogen content of vegetation leaves is to measure the sample in the field and use the chemical experiment method in the laboratory.Although the measurement precision is high,it needs a lot of manpower and material resources.Hyperspectral imaging technology combines imaging technology with hyperspectral technology,which not only plays the high spectral resolution,the band is continuous,but also makes use of the imaging technology to visualize and visualize the image.It is of great significance to accurately monitor nutrient content of vegetation.Taking apple orchard in Qixia,Yantai,Shandong as the research area,apple leaf samples were collected before and after May 2017 and the experimental data were measured.The image hyperspectral data of apple leaf samples were measured by imaging spectrometer,and the nitrogen content of apple leaf samples was measured in the laboratory.The hyperspectral response law of nitrogen content in apple leaves was obtained by classification of apple leaf samples.By further analysis of nitrogen and spectra,the correlation and sensitive wavelength of nitrogen content with the original spectrum of apple leaves and the sensitive band after the first order differential of the original spectral SG were obtained.The vegetation index related to nitrogen content in apple leaves was constructed and screened.On this basis,three prediction models of nitrogen content in apple leaves were established and the most predictive models were selected.The main research results are as follows:(1)The hyperspectral response law of nitrogen content in apple leaves was obtained.400-490 nm forms a low reflection area,490-560 nm forms a high reflection area,560-700 nm forms a low reflection area,and 700-750 nm has obvious red edge characteristics of vegetation,and the trend of 750-1000 nm is relatively gentle,forming high anti platform ejection.The spectral curves of apple leaves with different nitrogen content and different varieties were consistent.In the green light range,a green peak is formed at the wavelength of 550 nm due to the reflected green light.At green peak,the reflectance of spectral reflectance is sensitive to the nitrogen content of apple leaves,and the high spectral reflectance of apple leaf green peak(550nm)can be used to determine the nitrogen content of apple leaves qualitatively.At the green peak,the greater the spectral reflectance,the lower the nitrogen content of apple leaves;the smaller the spectral reflectance,the higher the nitrogen content of apple leaves.(2)By analyzing the correlation between the original spectral curve and the nitrogen content of apple leaves,two sensitive wavelengths,R550 and R723,were screened out.Through SG smoothing and first order differential transformation of spectral curves,9 sensitive bands are selected,which are SG-FDR403?SG-FDR469?SG-FDR525?SG-FDR566?SG-FDR650?SG-FDR696?SG-FDR781?SG-FDR851?SG-FDR933.By studying the vegetation index related to the nitrogen content of vegetation,and constructing three common vegetation indices of NDVI,DVI and RVI,5 vegetation indexes closely related to the nitrogen content of apple leaves were determined,which were TVI,TCI,NDVI(566,766)?RVI(566,766)?DVI(680,723),respectively.(3)BP neural network model,support vector machine regression model and random forest regression model were established for apple fruit leaf nitrogen content.Based on the original wavelength R550 and R723,SG smooth first order differential data SG-FDR403?SGFDR469?SG-FDR525?SG-FDR566?SG-FDR650?SG-FDR696?SG-FDR781?SG-FDR851?SG-FDR933,vegetation index TVI,TCI,NDVI(566,766)?RVI(566,766)?DVI(680,723),BP neural network model,support vector machine regression model and random forest regression model are established respectively.Compared with the results of the model,the nitrogen content support vector machine regression model SG-FDR-SVM based on the SG smoothing first order differential data is the best nitrogen prediction model.The estimated coefficient of determination is R2=0.724,the root mean square error is RMSE=1.94,and the relative error is RE=5.13%.