Effects of Pretreatment Methods and Bands Selection on Soil Nutrient Hyperspectral Evaluation

By using the laboratory reflectance spectra of 42 soil samples, the effects of spectral preprocessing and spectral band selection on predictive models of total kalian (TK) of soil were studied,. The results show that the first order derivative has the maximal correlation coefficient between all pretreatments, including Savizky-Golay smothing(S-G smothing), multiplication scatter correction(MSC), the first order derivative and the second order derivative. Compared with other bands selected and the pretreatment methods, the first order derivative which was used at 554-600 nm and 2205-2213 with multple linear regression(MLR) method combined with PC transformation could obtain a better TK prediction model, whose maximum determination coefficient(R2) is 0.774 and minimum root mean square error(RMSE) is 1.109.