Research On Parallel Processing Technologies Of Remote Sensing Data Based On CPU+GPU Under Single-computer

In order to Fufil the urgent need of the high-speed computing of large-scale data acquiredby remote sensing sensors，this thesis has studied the general computing technology based onGPU and some remote sensing data parallel processing algorithm in CPU and GPU underSingle-computer heterogeneous environment. The main work and innovation of this paper asfollows:1. The history and developing trend of parallel computing platform is briefly summarized.And the GPU’s hardware framework, software programming model, performance analysis model,optimization principle and basic strategies are discussed in detail, which provides the theoreticalbasis for fine granularity parallel processing through single GPU card. Moreover,theexperimental platforms used by this paper are given.2. An algorithm based on CUDA of image geometric processing is achieved. According tothe intensive algorithms characteristics of remote sensing image geometric processing, It isresearched to assign the work of geometric correction to large number of computing units inorder to achieve the purpose of parallel execution using CUDA-based parallel computingtechnology. By making full use of CUDA architecture of registers, constant memory, texturememory to minimize global storage use, the propose of full use of computing resources andeffectively reduce processing time is achieved.3. After researching SIFT feature detecting and matching based on GPU, a feature matchingmethod apply to remote sensing image of bedding face array by blocking its area is proposed.This method first adopt the robust RANSAC algorithm to estimate homography relationshipbetween two images to be matched and eliminate false matching points，then find overlappingarea between two images via the homographic matrix so as to achieve SIFT feature partitioneddetecting and matching on remote sensing image of bedding face array.4. A morphological LiDAR points cloud filtering method based on CUDA is achieved. Onthe basis of the traditional mathematical morphpolpgy LiDAR point cloud filtering, the methodadopt the parallel technique based on GPU to achieve the purpose of fast and reliable filtering inCPU and GPU heterogeneous environment.