Font Size: a A A

Research On The Construction Of Digital Elevation Model Data Based On Interferometric Phase Images

Posted on:2015-08-25Degree:DoctorType:Dissertation
Country:ChinaCandidate:H X HaoFull Text:PDF
GTID:1108330479479617Subject:Control Science and Engineering
Abstract/Summary:PDF Full Text Request
Digital Elevation Model(DEM) data are the basis of the construction and the simulation of the terrain in the battlefield. The DEM data with high precision play a very important role in strategic planning and aircraft navigation. As the development of the interference measurement, the interferometry phase images are widely used in the generation of the digital elevation model because of the high precision and effectiveness. Unfortunately, the interferometry phase images acquired from the sensors are often polluted by the noise and the observed phases are 2Ï€-periodic. Thus, it is highly significant to research on the generation of the digital elevation data which can be used in real applications from the interferometry observation.The main purpose of this thesis is to get the digital elevation data by phase denoising, phase unwrapping and phase to height conversion based on the interferometry phase images generated from the single look complex images, and to prepare the data for the construction of the virtual battlefield environment in military applications. For the denoising of the phase images, this thesis proposes interferometry phase denoising algorithms based on the sparse regression in the complex domain, including the denoising method based on the adaptive windowed Fourier transform and the denoising method base on the complexed valued dictionary learning. For the phase unwrapping problem, this thesis describes it as an optimization in Markov Random Fields and compares the effectiveness of the algorithms to solve the problem. At the end of the thesis, we research on the decomposition and comprehension of image blocks in the large scale images based on the real phase images from the Synthetic Aperture Radar, and we get the digital elevation data by geometric analysis. In all, the innovation of this thesis can be described as follows.1. This thesis researches on the principle of the digital elevation data measuring system and analyzes on the noise sources in the acquiring of the phase images. We study the effect of the noise and propose a simplified white Gaussian noisy model and an In SAR noisy model which is much more realistic. We also analyze the relationship between the two proposed models. Some performance indicators are presented to illustrate the competitiveness and effectiveness of the phase denoising and unwrapping algorithms. A simple fast denoising method named FLPA based on the local polynomial approximation is proposed. This algorithm is very fast and the essence of the algorithm is sparse representation. So the presence of the algorithm paves the way for the further research in interferometry phase denoising.2. We propose an adaptive window size selection method based on the two dimensional Fourier transform in the complex domain and make use of it in the phase image denoising to prevent the noise as much as possible and protect the details of the phase images. The algorithm, which is abbreviated as SAWFT, is based on the energy concentration to select the window size of every pixel. This algorithm acquires the sparse coding of the noisy phase image by hard thresholding the representation coefficiencies of decompression and gets the denoising results by multiplying the dictionary and the coding. The proposed algorithm can select the representation window sizes adaptively according to the phase image and the noise. The algorithm selects small window size in the fluctuated areas and the discontinuous areas in order to protect the details, while in the flat areas, the algorithm selects large window sizes to restrain the noise as much as possible. The experimental results show the effectiveness of the proposed denoising algorithm.3. This thesis also proposes fast algorithms to do the dictionary learning and coding in the complex domain, and they are used to denoise the phase images. The learning algorithm devides the whole phase image to overlapped patches and trains the adaptive dictionary by these patches. The phase image denoising method based on dictionary learning is abbreviated as Sp In PHASE in this thesis. The basis(dictionary) trained by the algorithm contains the details of the images and with little noise. So representing the patches in a sparse domain by the training set can filter the noise. We also get the conclusion that the estimation error is proportional to the sparsity level of the signal. That is the sparser of the representation, the more noise we can filter. From the experiments of the Sp In PHASE, the proposed algorithm can get higher peak signal to noise ratio than the state-of-the-art algorithms and protect the details of the discontinuous areas visually.4. We analyse the effect of the sampling and noise to phase unwrapping, and model the phase unwrapping problem by maximum a posteriori probability. Then the phase unwrapping problem is turned to the optimization of the Markov Random Field(MRF) problem. The algorithms to solve the MRF are researched and realized in phase unwrapping. We compare different algorithms to solve the MRFs in phase unwrapping and come up with a set of conclusions, including:(1) if the surface is continuous and contains little noise, multilable algorithms cost less time than binary lable algorithms;(2) PUMA and QPBOI can get better results than other algorithms for the discontinuous surfaces with much noise;(3) for the continuous surfaces, convex potential function and non-convex potential function can both be used to do phase unwrapping. It is better to choose the convex potential function because in this way the object function is convex. For the discontinuous surfaces, non-convex potential function can protect the details. Besides, this thesis also researches on the set of labels in phase unwrapping and gets the optimal object function and unwrapping results by selecting a proper unary parameter when the label set is not exact.5. This thesis realizes an experiment which follows the procedure of generating the digital elevation data based on the interferometry phase image which is distributed by European Space Agency. We take the strategy that dividing the large scaled image to small blocks and composing them together after processing in order to decrease the time and memory complexity. We analyze in theory why we can save time by decomposing the image to small blocks and processing them respectively. The thesis also studies how to get the geometric relationship between the absolute phases and the real digital elevation data by the parameters of the sensors and researches on the correction of the elevation data by the ground control points. Finally, the digital elevation model data can be generated by taking into account the distance of the pixels generated by the longitude and latitude of every point.The proposed algorithms in this thesis solve the problem of computing the digital elevation data with high precision which can be further used in the simulation of battle field, construction of virtual environment, and so on.
Keywords/Search Tags:Image denoising, Interferometry phase images, Phase estimation, Phase unwrapping, Sparse representation in the complex domain, Adaptive coding
PDF Full Text Request
Related items