Inversion Of Leaf Area Index And Chlorophyll Content For Winter Wheat Based On Wavelet Transform

To achieve precise and effective monitoring and management of wheat, timely and accurately monitoring wheat growth, nutritional status, fertilizer and information of plant diseases and insect pests is needed. As two biophysical and biochemical parameters in the process of winter wheat growth, the accurate and fast acquisition of leaf area index and chlorophyll content will help to make accurate diagnosis for the winter wheat plant diseases and insect pests of information and management control,but also has great significance in monitoring wheat growth and production evaluation,promoting agricultural informatization, the digital and precise construction.Affected by growth period, the environmental conditions and geographic conditions, etc, on the other hand, the lack of valid band information and easily affected by external disturbances, the existing retrieval methods of winter wheat biophysical and biochemical parameters are poor in model accuracy and universality. So building high-precision retrieval methods of leaf area index and canopy chlorophyll content is urgently needed to enhance the model adaptability at the canopy and remote sensing images. Recently, continuous wavelet transform and least squares support vector machine methods, for its powerful advantage, have been widely applied. However,combining the advantages of the two methods to improve leaf area index and canopy chlorophyll content inversion accuracy of northwest region is still relatively few. The paper takes the accurate acquire of leaf area index and canopy chlorophyll content as a starting point, a field joint experiment in Yangling zone is conduct, we regard model construction of leaf area index and canopy chlorophyll content and model localization as the main target based on the synchronous data of satellite and field, and improves the inversion accuracy of winter wheat leaf area index and canopy chlorophyll content from the ground monitoring to the wide range. This finally provide decision-making basis for large-scale agricultural production, regional guidance and precision agriculture construction, which has better theoretical and practical significance. The main contents and achievements of this paper are as follows:(1) The paper explains the significance of winter wheat leaf area index and canopy chlorophyll content in growth monitoring, production evaluation and construction of precision agriculture, discusses the advantages of CHRIS imaging spectrometer data,summarizes the main data source that used for physical and chemical parameters at home and abroad and the research progress of inversion method. At last, the shortages incurrent research are analyzed, on the basis, the main work, technical workflow chart as well as the outline of the paper is elaborated.(2) On the basis of introducing the general information of study area as well as the three core experiment sites, the experiment design and the samples distribution are described in detail. Then the measurement method of ground data, the characteristics of CHRIS imaging spectrometer data as well as imaging mode are elaborated. Finally, the samples of model calibration and validation are assigned.(3) Systematically elaborates the principle and advantages of continuous wavelet transform, least squares support vector machine, vegetation index, principal component analysis, stepwise linear regression and partial least squares regression method that adopted in building models of winter wheat leaf area index and canopy chlorophyll content. Under the condition of lacking spectral response function of CHRIS images,Gaussian function method is adopted to simulate spectral response function of CHRIS,then the spectral response function is used to resample the ground spectral. On the basis of analyzing the noise sources and removal method of CHRIS image, we adopt HDFclean method to remove the stripe noise on CHRIS image. Through comparing before and after the removal of stripe noise, result shows that the removing effect of noise is better. The experienced line correction method was used to conduct correction of the CHRIS image, through comparing with the spectra measured in the field at the same period, results show that the ELC spectra consistent with measured spectra in shape and character, and the absolute errors between the corrected and measured spectra at each band are all less than 5%.(4) With the canopy spectra resample in accordance with wavelength of CHRIS image, the correlation between leaf area index and canopy chlorophyll content of winter wheat and reflectance spectra as well as vegetation indexes is analyzed. The vegetation indices information, which is sensitive to canopy spectra and remote sensing image, are extracted. Through principal component analysis to canopy spectral, the first two principal components has already explained 98.275% useful information of the original spectral band, so the original information is instead by the two principal components for subsequent analysis. Through continuous wavelet transform to canopy spectral data,then make correlation analysis between the decomposed wavelet energy coefficient and winter wheat leaf area index and canopy chlorophyll content respectively, then we can obtain the correlation scalogram between wavelet power and winter wheat leaf area index and canopy chlorophyll content. Through analysis of the correlation coefficientscalogram, finally the 11 wavelet features that most sensitive to winter wheat leaf area index are extracted, which are(b12, scale1),(b12, scale2),(b11, scale4),(b1, scale2),(b1, scale3),(b10, scale5),(b9, scale6),(b8, scale7),(b7, scale8),(b6, scale9) and(b3,scale12) respectively, and 9 wavelet features are selected as the most sensitive features to winter wheat canopy chlorophyll content, which are(b12, scale1),(b16, scale1),(b11,scale4),(b9, scale6),(b8, scale7),(b7, scale8),(b1, scale2),(b1, scale3) and(b10, scale5)respectively.(5) On the basis of 11 wavelet features which are sensitive to winter wheat leaf area index, the paper uses the stepwise linear regression method to select wavelet features, and these features are introduced to least squares support vector machine, then inversion model of winter wheat leaf area index(WF-LS-SVM) are constructed based on continuous wavelet transform and least squares support vector machine. Through optimal selection with leaf area index models that based on vegetation index, principal component and reflectance band at the canopy, and then we conduct accuracy assessment to the screened model on remote sensing image, results show that the WF-LS-SVM model has the highest prediction accuracy on remote sensing image, with the R2 and RMSE 0.55 and 0.67 respectively, next is the WF-SLR model(R2=0.45,RMSE=0.78), followed by are the models of R-LS-SVM, PC-LS-SVM and MSR-LAI,accuracy of the R-PLSR model is the lowest(R2=0.22, RMSE=1.22). Comparing with the vegetation index models, the WF-LS-SVM model has a better accuracy and better adaptability to growth period, which properly solved the problem of lower model accuracy and stability of vegetation index model using few band information, easily affected by factors such as vegetation types and geographical conditions, the WF-LS-SVM model has certain adaptability at canopy and remote sensing image.Finally, with the help of synchronous CHRIS hyperspectral images, we achieve the remote sensing mapping of winter wheat leaf area index in Yangling and surrounding areas.(6) Continuous wavelet transform method can effectively capture some spectral information that sensitive to vegetation biophysical and biochemical parameters, the least squares support vector machine method has a powerful advantage in model regression. Winter wheat canopy chlorophyll content model that based on multiple wavelet features is established combining the advantages of the two methods.Through optimal selection with winter wheat canopy chlorophyll content models at canopy, and then we conduct the accuracy evaluation to these models on remote sensingimage. Results suggest that the WF-LS-SVM model established combining multiple wavelet feature with least squares support vector machine method has the best accuracy(R2=0.58, RMSE=0.40). Results suggest that the continuous wavelet transform method can effectively extract wavelet features that more sensitive and robustness to winter wheat canopy chlorophyll content, and comparing with band reflectance, wavelet features enhance the correlation with winter wheat canopy chlorophyll content. Thus the WF-LS-SVM model, combining multiply wavelet features with least squares support vector machine, has the best accuracy, and it enhances the adaptability to winter wheat canopy chlorophyll content at jointing and grain filling stages. Finally, the paper promotes the WF-LS-SVM model to Yangling experimental area, and achieves ground-air integration monitoring of winter wheat canopy chlorophyll content in Yangling and the surrounding area...