Remote Sensing And Change Detection Research Based On Compressive Sensing

It's very difficult to sample change area data directionally due to uncertainty and unpredictability of sampling objects. Conventional data collecting of entire zone leads to repeat work towards unchanged areas so that capital and resource are wasted. Compressive sensing (CS) theory can be used properly in directional change detection and incremental updating research of urban remote sensing image by virtue of the sparsity hypothesis of change area. It has been demonstrated in our early experiments that change area can be reconstructed losslessly by2-3times as much as its data based on the difference of unlike temporal CS measurement values. Two-dimensional CS model is proposed in order to use change areas' prior information such as spatio-temporal continuity and structure based on data acquisition features of remote sensing images. Our research starts from change detection and incremental updating to pave new thought for reconstruction of remote sensing image, spreading it in the general mechanism research. Hence, the research of the project has great scientific value and broad engineering application prospect. The main contents of this paper are as follows:(1) Partial hadamard matrices have strong signal recovery ability. The main influencing factors that lead to partial hadamard matrices' excellent behavior are summarized by the comparative analysis of statistical parameters of partial hadamard matrices and Gauss matrices. The origin, formation and improvement of the optimization algorithm of Gauss matrices are described in detail. The validation and generality of optimization algorithm are verified by Gauss matrices of different size. Optimal matrices'applicability to different sparse signal reconstruction algorithms is determined. At last, preliminary theory experiment analysis of optimal matrices'performance is conducted. Research shows that optimal matrices are as much as partial hadamard matrices more in the signal recovery ability. The research results provide a new idea and method for the analysis, design and optimization of measurement matrices.(2) Current optimization of measurement matrix of compressive sensing is optimization beforehand by use of the same matrix in measurement and reconstruction stages. Transition matrix and optimization algorithm mainly based on row transformation are proposed to separate measurement matrix and reconstruction matrix of compressive sensing.0-1sparse matrix of single-pixel camera is adopted during measurement; approximate matrix is adopted during reconstruction in the paper. It's a kind of afterwards optimization method of measurement data and measurement matrix. It's different from traditional thinking. Theory analysis and experiment results demonstrate that characteristics of optimal matrix are better than circulant sparse matrix, approximate matrix and optimal matrix have similar characteristics. The research results reduce the difficulty of engineering design and implementation of measurement matrix.(3) One-dimensional compressive sensing measurement data based on Gaussian measurement matrix not only well retain sparse signal's energy, also inherited sparse signal's direction information. But in the one-dimensional compression sensing model, direction information can not be applied to sparse signal reconstruction and examination. Two-dimensional compressive sensing model is proposed based on sparse features of remote sensing image change area. And by use of energy and direction information, sparse signal reconstruction algorithm (2D0MP) is constructed based on two-dimensional compressed sensing. Theoretical analysis and experimental results demonstrate signal reconstruction ability of2D0MP algorithm is stronger. Meanwhile, the concepts of directional remote sensing and directional change are put forward based on the fact that very little measurement data are required to recovery sparse signal by compressive sensing.