SAR Image Compression And Reconstruction Algorithm Based On Compressed Sensing

Compressed sensing theory as a new data acquisition technology, it was concerned bymany domestic and foreign scholars once it was put forward. It broke the shackles of theNyquist sampling theorem, made the fast compression high of resolution signal possible. Inthis paper, based on the classic algorithms of compressed sensing has been analyzed, aimingat the construction of measurement matrix and reconstruction algorithm improvement, we putforward our views. And the improved compression sensing algorithm is applied to the SARimage compression and reconstruction, the experimental results show the effectiveness of thealgorithm. The main work includes the following aspects:1. The contour wave (Contourlet) transform was used to represent SAR image as sparsesignal. Because the low frequency sub-band coefficients contain the main information and donot meet the sparse conditions, so taking full sample. Then, according to the different texturefeatures of the high frequency sub-band coefficients, taking sample respectively on the rowsor columns, then realizes SAR image sparse representation finally.2. Gaussian random matrix is chosen as the observation matrix. However, the matrix hasits own denseness, which resulting in the very large amount of calculation. Its approximateQR decomposition makes the minimum singular value of matrices increase.Thereby, itimproves the linear correlation of the matrix and reduces the computational complexity.3. Sparsity adaptive matching pursuit (SAMP) reconstruction algorithm can achieveaccurate reconstruction in the case of unknown sparsity. But the setting of its fixed step size isprone to lead to overestimation and underestimation. On this point, this algorithm combinesthe SP algorithm initial sparsity estimation methods with ROMP algorithm regularizationprocess, to control the number of iterations effectively, achieving a fast and accuratereconstruction.4. SAR image despeckling. The inherent speckle noise of SAR image seriously affectedthe visibility of image. Therefore, after the Nonsubsampled Contourlet Transform(NSCT) ofSAR image, we use the maximum a posteriori estimation (MAP) method to adjust thehigh-frequency coefficients, and use improved smoothed L0norm (SL0) algorithm for preciserecovery. The experimental results show that this method greatly improves the visibility of theimage.