Research And Application On SAR Image Mosaic

Synthetic aperture radar (SAR) was born in1950s. It can work all-weather, all day,and the SAR image shows its unique advantages compared to the optical image. Thedevelopment of spaceborne, airborne and Unmanned Aerial Vehicle (UAV) SAR imagingtechnology has greatly pushed forward the application of SAR in national economic andnational defense construction.Constrained by the imaging equipment and technology, image perspective can’tmeet the practical needs. In order to expand the scope of SAR imagery scene, thesub-aperture images with overlapping area are usually made into SAR image with awhole scene, namely the SAR image mosaic technology. Therefore, the research andapplication of SAR image mosaic technology are very important.In the paper, SAR image mosaic technology was studied, and the technology wasused to stitch a group of SAR image sequence and the SAR image sequence mosaic in2-Dimensional space was realized.The main work of this paper is as follows:1.The basic characteristics of the SAR image were introduced, and a comprehensiveanalysis of each step of the SAR image mosaic technology was made.2.At the stage of SAR image pre-processing, the mixed Fourier-Wavelet de-noisingmethod was used to deal with the coherent speckle noise in the SAR images and thecombine of the top-hat and bottom-hat filter and contrast enhancement method wasapplied on the uneven illumination in the SAR images. The experimental resultsdemonstrate the validity of the method.3.At the stage of SAR image registration, the SAR image registration algorithmbased on the improved PCA-SIFT was used. At first, the feature points in the SAR imagewhich were after the image pre-processing were extracted by using the feature pointextraction method based on the improved PCA-SIFT; Then, the K-D tree was built, thematching points were sought by using the BBF query mechanism and Mahalanobisdistance; And the affine transformation model was used to calculate the transformationmatrix; Finally, the improved RANSAC method was used to eliminate mismatch of thepoints and the transformation matrix was further optimized combining with the bilinearinterpolation method.4.At the stage of SAR image mosaic, a SAR image mosaic method based on a full binary tree model was proposed, and it was used to complete the SAR images mosaic.First, according the principle of “from the bottom up, from left to right”, the stitchingSAR image sequence was taken as the leaf node to generate a full binary tree model;Then with a layer of the binary tree model for a unit, according to the principle of “thebottom up” and combining the thought of “parallel” and “recursive”, the local SARimage mosaic was completed. At last, based on the transformation matrix and resamplingtechnology, the SAR image sequence alignment was completed, until finally a wholeSAR image was generated. Comparing with the other global motion model, the proposedmethod is of high efficiency and easy to calculate, and from the view of subjectivity andobjectivity, the final whole SAR image generated by the proposed method is more ideal.