Research On SAR Image Parallel Processing Technology

With the evolving of the Synthetic Aperture Radar (SAR) imaging technology, and theincreasing of the amount of SAR image data, the mosaics time-consuming has grownrapidly. The conventional serial mosaics technology of SAR image has been unable toadapt to the requirements of areas which has high real-time requirements, such as Defenseand military disaster warning. In this paper, we used the Cluster Computing technology toimprove the rapid of SAR image mosaics. And improved the efficiency of the SAR imagemosaics effectively under the premise of guaranteed the treatment effect.Firstly, the characteristics of the parallel processing system and the SAR imagemosaics technology have been introduced, and the installation and configuration of theparallel system have been completed, which included the select of hardware and software.The installation and configuration of the parallel processing system was very complex,which included the configuration of cluster system node machine, the installation andconfiguration of MPI, the configuration of OpenMP programming environment, and theconfiguration of application software development platform.Secondly, we have completed the pre-treatment operation of SAR image whichincluded the noise removal and the uneven illumination correcting based on the using ofthe master-slave communication modes of parallel system. The SAR image pre-treatmentwork has two methods of the noise removal method based on the TV model and the imageenhancement method based on homomorphic filtering model. Respectively, the parallelpre-treatment and comparative analysis of parallel processing effect had been done byusing different SAR image size of1M,4M and16M on the different cluster node numberof2,4and8environment. And OpenMP multi-threaded had been used to parallel improvethe noise suppression method based on the TV model, and comparative experiments ofparallel processing effect had been done by using different size of SAR image in1-nodeenvironment, and4-node environment. The experimental results show that, parallelpre-processing method can reduce the pretreatment time of the SAR image effectively, andimproving the pre-treatment efficiency.Thirdly, a parallel registration method of SAR images based on improved SIFTalgorithm have been proposed, and which been combined with the OpenMP multi-threadedparallel programming to complete the parallel characteristics extraction of SAR imagesbased on the registration method. In this method, with the contrapose of the characteristicsof the SIFT algorithm, we added a pre-screening after the feature point detection step, to reducing the amount of data in the follow-up steps, according to reduce thetime-consuming of feature extraction under the premise of guaranteeing the featureextraction effect.Finally, we proposed a complete binary tree model based segmentation that increasein the traditional complete binary tree model with the idea of image segmentationprocessing, and through which to parallel mosaics processing of SAR image. The methodin this paper is different from the traditional parallel processing. In this paper we useddifferent image number of SAR image sequence an example, to complete the parallelmosaics contrast experimental of the proposed method and traditional methods. The resultsshow that the proposed method is generally superior to the traditional method on parallelefficiency. Compared with the traditional method, the proposed method is more efficientand comprehensiveness.