Successive Projection Method Applied To The Wavelength Selection Of Soil Organic Matter With NIR Spectroscopy

Organic matter is an important soil testing indicator, existing measurement methodsusually need chemical reagents, tedious and time-consuming operation; rapid analysis ofsoil organic matter using near infrared (NIR) spectroscopy has significance and applicationvalue. This paper focuses on the study of soil organic matter NIR spectroscopy wavelengthselection and model optimization.Reliable analytical models were established by dividing the calibration, prediction andtest sets based on the randomness and stability considerations. NIR analysis model of soilorganic matter were established by using partial least squares (PLS) method and continuousprojection-multiple linear regression (SPA-MLR), respectively. Five conventional divisionsof the band: visible (400-780nm), short-wave NIR (780-1100nm), long-wave NIR(1100-2498nm), NIR (780-2498nm) and full-spectrum (400-2498nm), each PLS modelwas established, respectively. By comparison, the PLS model of the long-wave NIR has thebest model effectiveness and stability, the optimal number of PLS factor was6, the standardroot mean square error of prediction (V-SEP) and the correlation coefficient (V-RP) of testwere0.243,0.931, respectively;7wavelengths in the NIR region were selected through theSPA-MLR method, they were922,1018,1132,1930,2142,2206,2498(nm), the V-SEPand V-RP, respectively, were0.248and0.928.The prediction effect of the combination of MLR model of the discrete wavelengthsclose to the PLS model of the long-wave NIR region, but the model will greatly reduce thecomplexity of the spectroscopic system, which provides a valuable reference for the designof soil dedicated spectrometer.