Inversion Of The Wheat Biophysical And Biochemical Parameters Based On Hyperspectral Remote Sensing

It was the basis of agricultural production to obtain the biophysical and biochemical information of crops accurately and quickly. However, the traditional chemical method for crop parameters acquisition has the disadvantages of time-consuming, only for point-source information and difficult to acquire information at the macro scale, and that greatly influenced the comprehensiveness, timeliness and objectivity of agronomic decision-making. The improvement of hyperspectral technology remote sensing and its great advantage in quantitative analysis brought a new opportunity to solve the problem, according the informed research, the new methods which combined the hyperspectra remote sensing technology and field spectroscopy technology were improved and developed to retrieve the wheat nitrogen content, chlorophyll content, water content and leaf area index accurately and rapidly in this paper, and then provided scientific support for agricultural production. The main results of this study are as following:1. The pretreatment methods of hyperspectral remote sensing image and field spectra were improved. Firstly, the field spectra were pretreated by the method of wavelet denoising combined with first derivative to eliminate the background information and random noise of the original spectra. And the problems that the field spectra mixed with random noise and background information were solved. Secondly, the hyperspectral image was denoised with minimum noise fraction algorithm, and the atmospheric correction method was optimized by quick atmospheric correction combined with experience linear correction. Then the image with the true surface reflectance was acquired. The improvement of the field spectra and hyperspectral image's pretreatment methods laid a good foundation for wheat biophysical and biochemical parameters'inversion.2. The advantage and disadvantage and the adaptation range of the target area extraction methods included vegetation indices extraction, classification extraction and endmember spectral matching mapping extraction were systematical analyzed. The result were as following: the method of vegetation indices extraction has the advantage of simply to operate and fastly to calculate, but it was difficult to distinguish different vegetation type, so it was suitable for single vegetation type extracting or no need to distinguish different vegetation types. The method of classification extraction has the disadvantage of the training samples were selected subjectively by operator. However, it had a better interaction and the training samples could be adjusted. And it could divide the feature type according to specific requirements and got the method of feature type dividing which was suitable for target area extraction. Therefore, this method was suitable for mixed area extraction. The method of endmember spectral matching extraction has the advantage of accuracy and objectivity, but it was poor for mixed area extraction, and suitable for pure land cover type extraction.3. Canopy leaf nitrogen content was retrieved with the hyperspectral index FD-NDNI (first derivative normalized difference nitrogen index) which was put forward in this paper. And the inversion model was optimized by Least squares support vector regression (LS-SVR) algorithm. The hyperspectral index FD-NDNI was the normalized value of the red edge and blue edge's steepness of wheat canopy reflectance spectra, and it could represent the nitrogen content of wheat canopy sensitively. The calibration R-square (C-R2) and prediction R-square (P-R2) of the inversion model which built by index FD-NDNI were 0.846 and 0.838, respectively, higher than the inversion model which built by the commonly used indices such as mNDVI705,mSR705 and NDVI705. Further analysis showed that the correlation between index FD-NDNI and leaf area index was the lowest in all the hyperspectral indices (with the correlation coefficient 0.67), and indicated it could prevent the plant canopy density to affect the estimation of nitrogen content most effectively. The nitrogen content information of wheat canopy was space quantitatively expressed in OMIS image based on the inversion model, and then the inversion and measured values was compared by the method of regression fitting. The R-square and RMSE of the fitting model were 0.676 and 1.715, respectively, indicated the similarity between of inversion value and measured value was high. The result indicated that FD-NDNI was an optimal hyperspectral index for nitrogen content estimation of wheat canopy, so a new method for wheat canopy nitrogen content inversion was provided.4. Eighteen kinds of hyperspectral indices were comparative analyzed. The index REP, which could respond wheat canopy chlorophyll content sensitively, was selected, and the physical interpretation of that was discussed. The inversion model of wheat canopy chlorophyll content was established using the index REP and then was optimized by LS-SVR algorithm. Analysis showed that the accuracy of the inversion model which established with REP was highest in all hyperspectra indices (C-R2=0.751, P-R2=0.722), and the goodness of fit between remote sensing mapping values and ground measured values was high (R2=0.676). It indicated that REP was an optimal index for wheat canopy chlorophyll inversion.5. The new hyperspectral indices (FD730-955, FD730-1145 and FD730.1330)for estimating water content of wheat canopy was put forward, and the physical interpretation of that were researched. The inversion of wheat canopy water content was achieved using the optimal index FD730-955, and the inversion model was optimized by LS-SVR algorithm. Analysis showed that the C-R2 and P-R2 of the inversion model which established by index FD730-955 were 0.797 and 0.820, respectively, higher than that which established by the commonly used indices such as WBI, NDWI and MSI. And the goodness of fit between remote sensing mapping result and ground measured result was high too (The R2 of the fitted model reached 0.676). It indicated that FD730-955 could represent the canopy water content nicely, and it was an optimal index for water content inversion.6. Eighteen kinds of hyperspectral indices were comparative analyzed. The index OSAVI, which could respond wheat leaf area index (LAI) sensitively, was selected, and the physical interpretation of that was discussed. The inversion model of wheat LAI was established using the index OSAVI and then was optimized by LS-SVR algorithm. Analysis showed that the accuracy of the inversion model which established with OSAVI was highest in all hyperspectra indices (C-R2=0.839, P-R2 =0.836), and the goodness of fit between remote sensing mapping values and ground measured values was high (R2=0.756). It indicated that OSAVI was an optimal index for wheat LAI inversion.