| In this study, the hyperspectral data of four different kinds of leaves including healthy mature, zinc
deficiency mature, leaf-miner inflicted mature and healthy young leaves, which have different pigment content levels,
were collected in a pomelo agricultural garden in Meizhou city. It was attempting to investigate the best hyperspectral
remote sensing models of pigment contents estimation of pomelo leaves under different environmental stresses. The
total chlorophyll and carotenoid content of the leaf samples were determined by biochemical methods. On the basis of analyzing the correlation between pigment content value and original spectral reflectance, differential spectrum
and hyperspectral characteristic parameters, bands and hyperspectral characteristic parameters with extremely good
correlation （P < 0.01）with pomelo pigment content in each category were selected. The linear, logarithmic and
exponential models of single variable regression and multivariate linear stepwise regression were employed to estimate
the content of two pigments in the four kinds of pomelo leaves. The estimation models established by multivariate
linear stepwise regression have the highest precision. The modeling precision of chlorophyll and carotenoid of healthy
mature pomelo leaves is 0.850 and 0.705, and the testing accuracy is 0.754 and 0.606, respectively. The modeling
precision of two pigments for zinc deficiency mature leaves is 0.895 and 0.904 and the testing accuracy is 0.932 and
0.908, respectively. The modeling precision of chlorophyll for leaf-miner inflicted mature leaves is 0.738 and the
testing accuracy is 0.834, respectively. The modeling precision of two pigments for healthy young leaves is 0.911 and
0.897, and the testing accuracy is 0.898 and 0.944, respectively. It is suggested that the multivariate linear stepwise
regression models be used to estimate the chlorophyll and carotenoid contents of pomelo leaves under different stresses
based on hyperspectral remote sensing. |
