Inversion Feature Based Hyperspectral Reed Chlorophyll Content Wavelet Transform

At present, remote sensing technology has become an important means about a plant growth condition monitoring of large area, the application about hyperspectral remote sensing in vegetation growth state diagnosis gradually become the focus and attention by people to study. How to obtain the vegetation physiological state information more quickly and accurately is very important for research the ecological environment effect.The reed is a typical wetland vegetation, monitoring the growth of reed have important meaning to research the wetland ecological environment effect, at the same time, to establish a rapid, accurate and low cost method of the spectral analysis, in order to replace a time-consuming and high cost method of plant chemical analysis has also important significance.The South Wetland in the Olympic Park was selected as the research area. The reed in different wetland purification system were treated by field measurement and transplanted to laboratory by nitrogen gradient process. Then both indoor and field scale reed hyperspectral data were gained as the research data. These two kinds of spectral data were transformed by first derivative and one dimensional wavelet decomposition(2,4,8,16,32,64scales) as input data type. And the reed chlorophyll content inversion models were builded using stepwise regression method. Then the model was optimized by the nonparametric test and multiple factor variance analysis and tested by the leave-one-out cross validation correlation factor R2CV and root-mean-square-error RMSECV. This article research conclusion is as follows:(1) The single factor optimization results of inversion modle based on indoor and outdoor reed leaf chlorophyll spectral data showed that the wavelet decomposition scale is the most important factor to affect the precision of model, followed by the model input data type, and wavelet type has no significant effect for model precision. Each factor interaction analysis results showed that the interaction between the input data type and decomposition scale is a significant effect on the precision of model, the others factors interaction has no influence.(2) The sensitive wave positions of inversion model based on indoor and outdoor reed leaf spectral data are different. For indoor leaves, wavelength of higher correlation between chlorophyll content and reflectance spectra waveband were mainly located in the blue and red waveband in the visible light range, and wavelength of higher correlation between chlorophyll content and its derivative spectral waveband were mainly located in the blue,green and red edge waveband. For the original reflectance spectra of outdoor leaves, wavelength of higher correlation between chlorophyll content and waveband were mainly located in the green and red waveband, and wavelength of higher correlation between its derivative spectral waveband and chlorophyll content were located in the red edge waveband and near-infrared light range. The sensitive wave positions of inversion model based on these two kinds of data were mainly located in the red edge waveband. Then we can find that the red edge band is stable wavelength range to affect the inversion of chlorophyll.(3) The research results of the uncertainty factors of the inversion model based on indoor and outdoor reed leaf spectral data showed that the noise information were amplified when the original reflectance spectra was transformed by first derivative, and the influence of the uncertainty for precision of model were not amplified by the derivative spectrum transformation and wavelet coefficients transformation. And the influence of the uncertainty factors that based on derivative specture of outdoor data were less than indoor data.(4) For indoor spectral data, the R2CV and RMSEcv of the model that using biorl.5wave coefficient of derivative spectrum in the scale4were0.879and4.776mg/cm2, and it is a optimal model. For outdoor spectral data, the R2CV and RMSECV of the model that using coif2wave coefficient of derivative spectrum in the scale4were0.8202and0.354mg/g, and it is the optimal.The research achievements of this paper can provide some reference and powerful science basis not only for the wetland vegetation growth status detection of remote sensing but also for the urban wetland monitoring and management.