RaPid Detection Of Nitrogen Content In Rubber Leaves Based On Hyperspectra Data

Nitrogen is one of the most important nutrients of rubber tree (Hevea brasiliensis), which is closely related with the growth and yield of rubber tree. Real-time fast and accurate determination of nitrogen content of rubber leaf is critical for prediction of nutrition diagnosis and prescription of nitrogen topdressing. In this paper, prediction model was build by analyzing nitrogen spectrum characteristics, and realized fast and accurate determinating of nitrogen content of rubber leaf,provided a theoretical basis for nitrogen nutrition remote sensing monitoringDeterminate total nitrogen content of rubber leaf. Spectral information of rubber leaf have been collected by American ASD spectrometer and high-density vegetation probe. By analyzing the spectral curve, knew the reflection-absorption and coding characteristics; By analyzing the relationship of the reflection spectrum and the nitrogen content, defined the sensitive band of nitrogen, defined the best ratio vegetation index for estimating nitrogen content of rubber leaf; Extracted trilateral characteristics of rubber leaf by data of derivative spectrum, analyzed the relationship of characteristics and nitrogen content; Used spectrum data(reflectance,absorbance,first derivative)as information source, selected modeling samples by different methods, builded models(PLSR) and evaluated it.Result:(1) Rubber leaf has obviously spectral characteristics "peak" and "valley" which is different from deionized water and soil; Reflection spectrum in 465nm,538nm,663nm,730nm,700-1300nm and nitrogen content has a high correlation, extremely significant level; The correlation between reflectivity in 730nm and nitrogen content is extremely, the models'coefficient of correlation (r) reached 0.8422; Defined the best ratio vegetation index RVI (R1298/R740) for estimating nitrogen content of rubber leaf, builded linear equation, y= 8.0797x-5.9582, R2=0.7396; (2) Differential spectrum in 557nm,635nm,1409nm,1744nm,2320nm and nitrogen content has a high correlation, differential spectrum in 714nm and nitrogen content is extremely, R2 is 0.6721; In "Trilateral" characteristics, DR,DB,SDY,SDB,DR and nitrogen content has a high correlation, builded model with red edge integral area of variable and nitrogen content, R2 is 0.6389, correlation of other variable is not very significantly; (3) Derivative spectrum is help for lower the number of composition, reduce the computational, facilitate the best fitting of the model; (4)Derivative spectrum model is more stable to the reflection-absorption spectrum models, and the predictive power is strengthen; The predictive power of the model builded by PCA-grid method is strength than the models builded by Random and Content grads; The reflection-absorption derivative spectrum models builded by PCA-grid method has a high correlation in self-correcting, were 0.9807,0.9591, prediction samples R2 is 0.9599,0.9492, prediction standard deviation were 0.0978,0.1061,which can accuratly predict the nitrogen content of rubber leaf.