| Both modeling and model-based inversion are important for quantitative remote sensing. Hundreds of models related to vegetation and radiation have been established during the past decade years. However, the model-based inversion caused the attentions of scientists only in recent years. Compared to modeling, model-based inversion is still in the stage of exploration. Lots of difficulties exist in the application of a priori information, inverse strategy and inverse algorithm. The appearance of hyperspectral and multiangular remote sensor enhanced the exploration means, and provided us more spectral and spatial dimension information than before. How to utilize these information to solve the problems faced in quantitative remote sensing to make remote sensing really enter the time of quantification is an arduous and urgent task for remote sensing scientists. Remote sensing inversion is paid more and more attentions in recent years. In a series of international study projections, such as IGBP, WCRP and EOS, remote sensing inversion is included as a focal point of study. In China, scientists, represented by Prof. Li X.W., not only have made great achievements in forward modeling, but also realized the importance of remote sensing inversion and studied it as a special direction early. They've done lots of creative work for remote sensing inversion. This study arises from the above background. Its main aim is to perfect current inverse theory. On this basis, exploring the method and feasibility to extract terrestrial parameters from hyperspectral and multiangular data. On the inverse problem, we analyzed the error source of model-based inversion, and pointed out the effect of a priori information, merit function, inverse algorithm and strategy on the inverse result. On this basis, Improving the Bayes inversion and the USM-based multistage inverse strategy to make them adapt to uncertain inversion. At the same time, we introduced the real-valued Genetic Algorithm (GA) to remote sensing inversion, and compared it with deterministic searching method. The result shows that GA functions well for the nonlinear inverse problem. In this dissertation, we put forward the idea and method to combine hyperspectral and multiangular data together to extract terrestrial parameters, and tried to inverse component signatures and structure parameters from Boreas data using GOMS model and linear spectral unmixing model. The dissertation has following special points: 1. Putting forward the concept of broad sense and narrow sense remote sensing inversion and the concept of subpixel unmixing between classes and within classes. 2. Analyzing the error source of remote sensing inversion, and providing the corresponding solution scheme. 3. Improving the USM-based multistage inverse strategy. Using the method to vary insensitivity parameters according to their priori distributions instead of fixing them to priori values to acquire the uncertainties of parameters to be inverted. 4. Suggesting paying some attentions to the constraint conditions for uncertain inversion to make them suitable to circumstances where models or observed data are not certain enough. 5. Introducing real-valued GA to remote sensing model inversion, and improving the objectivity and global convergence of traditional GA from some aspects, such as the initial population, the crossover rate, etc. 6. Comparing GA with general deterministic searching method, analyzing the efficiency and deficiency of GA, and approving the validity of GA in resolving the problem of nonlinear model inversion. On this basis, putting forward the mixed GA method. 7. Presenting the idea to extract terrestrial parameters from hyperspectral and multiangular data together, and studying its validity. 8. Studying the method to acquire and express a priori information, and probe the possibility to extract a priori information from remote sensing images directly. 9. Deducing the formula to calculate crown cover projection(CCP) based on Geometrical-Optical model.
|
PDF Full Text Request