| Citrus phosphorus content plays an important role in citrus physiological and ecological process. The quickness, accuracy and nondestructive determination of citrus phosphorus content provide references for a reasonable phosphate fertilizer, which can improve the yield and quality of citrus. The rapid development of hyperspectral technology makes it possible to measure citrus phosphorus nondestructively. Field experiments were conducted on 117 planted Luogang citrus trees in the Crab Village of Guangzhou and 234 pairs of data sample were collected in two different phenological periods, respectively, bloom period and picking period. Hyperspectral reflection data was used as highdimensional vector description, and phosphorus content measured by chemical method as true label to model and to predict the phosphorus content of citrus leaves. A regression analysis algorithm, least squares support vector regression were used to model and predict with the spectral reflectance data denoised by wavelet transform. Validation and calibration sets were used to evaluate the predictive performance of the determination model. The model achieved coefficient of determination of 0.907 and 0.953, the mean square error of 0.004 and 0.002 as well as the mean relative error of 2.76%and 1.77%. The experimental results showed that it is an effective way to predict phosphorus content based on hyperspectral reflection data which is denoised by wavelet transform. |
