Nitrogen Diagnosis Study On The Rubber Tree With Visible Light/Near Infrared Spectrum

Rubber trees are the significant production source of natural rubber in our nation and the nitrogen content is the important nutritive index of them. For quickly and non-destructive detection of the nitrogen content of the mature rubber tree leaves, the judgment about nutritive index of the trees can be determined. The quickly detection algorithms and models of the rubber trees nitrogen content with visible/near infrared spectrum were established in this study and the major studies are as follows:(1) Feature extraction methods of the spectrum datum of the rubber tree leaves were studied and the regression models were established. Extraction method of the spectrum feature datum of the rubber tree leaves based on SCARS(Stability Competitive Adaptive Reweighted Sampling) was proposed after analyzing the characteristics of the spectrum datum. Then, this method was combined respectively with non-linear regression models:BPNN and LSSVM to establish the quantitative regression models: SCARS-LSSVM and SCARS-BPNN.(2) Common models were compared with the quantitative regression models propsed in this study. Nitrogen content were predicted by SCARS-LSSVM and SCARS-BPNN on the band of visible light(450nm?780nm), near infrared(781nm-2400nm) and visible light/infrared(450nm?2400nm) respectively. Six common models: PCA-BPNN,PCA-PLS,PCA-LSSVM,SPA-BPNN,SPA-PLS and SPA-LSSVM were established to compare with them. The results show that the prediction effect of the SCARS-LSSVM was superior to the six common models. The correlation coefficient prediction of SCARS-LSSVM (Rp)is 0.9568 and the root mean square error prediction of it (RMSEP) is 0.1168.(3) The major feature band of the spectrum detection of the rubber tree leaves was confirmed. Datum extraction method: SCARS, SPA and PCA were compared and the cumulative variance contribution rates of the 3 principal component in PCA were above 0.90. However, the 3 principal components were not enough to represent the major feature of this set of datum because the samples without nitrogen stress treatment resulted that the spectrum datum were scattered. On the band of 450nm-2400nm, 27 points were extracted by SPA and 42 points were extracted by SCARS. Analyzing several bands of the higher correlation coefficient absolute value of spectral reflectance data and nitrogen content found: on the band of 700nm?1380nm, 8 points were extracted by SPA and 11 points were extracted by SCARS. On the band of 1800nm-1850nm, 2 points was extracted by SPA and 5 points were extracted by SCARS. On the band of 2150nm-2250nm, no point was extracted by SPA and 6 points were extracted by SCARS. Therefore, the band of 719nm?1342nm,1800nm?1850nm,2150nm?2250nm is the sensitive band of the nitrogen spectrum of rubber tree leaves.