Study On Architecture Of Cooperative Processing For Remote Sensing Images And Its Application On Photogrammetry

Nowadays, the types and amount of remote sensing images are expanding rapidly, it is of great significance and practical application value to establish a rapid, efficient and extensible remote sensing image processing platform combining with high performance computing technology and digital photogrammetric processing technology. This dissertation makes a comprehensive study on the architecture and photogrammetric application of high-performance remote sensing image processing system, of which the key technologies, such as cooperative processing service grid building, scientific workflow analysis and management, multiple baseline stereo image matching and so on, are deeply discussed and studied. The main achievements and innovations are summarized as follows:1. Through analyzing the characteristics and core technologies of the new generation digital photogrammetry parallel processing system, current research status and technology of the system is comprehensively summarized. And of the system, technical problems which have to be further developed are sorted out.2. A flexible and extensible software and hardware architecture, which is suit for remote sensing data processing characteristics and requirements, is proposed from the point of remote sensing image processing capacity. Based on the open architecture design method, a cooperative processing architecture for remote sensing images is constructed. Deployment scenario and deployment structure of the system is clearly made, and operation mode, operation process and data stream of the system is designed.3. A flexible, dynamic and convenient cooperative processing grid for remote sensing images is constructed. Based on Condor, an overall structure of the grid is designed. Meanwhile, a service-based grid gateway is constructed, and the methods and strategies of dynamic dispatch are proposed under grid environment.4. A scientific workflow structure with high performance and dynamic interaction is proposed. Both development model of workflow system related elements and workflow scheduling process under grid environment are designed, and the auto-running of data and processing tasks for remote sensing images can actualize through customizing processing pipeline.5. Algorithms of multi baseline stereo matching and ortho-rectification are studied and optimized. A pyramid semi-global matching(P-SGM) algorithm is designed. And combining P-SGM algorithm, geometrically constrained cross correlation(GC3) model and multi-view vertical line locus(MVLL) model, a new GC3-P-SGM, MVLL-P-SGM multiple baseline stereo matching model is proposed. Meanwhile, in order to improve the processing speed, a coarse granularity GPU parallel computing scheme for dense matching tasks and a fine granularity GPU optimizing scheme for resampling tasks are designed.6. According to the characteristics of photogrammetric processing objects and algorithms, the photogrammetric cooperative processing pattern and its inner parallel mode are determined. And general parallel interfaces are proposed to providing a convenient way to parallel transform and integrate more photogrammetric processing algorithms.7. Photogrammetric processing algorithms are seamlessly integrated with workflow management and remote sensing image processing grid platform. And a mixed cluster cooperative processing system, which is suit for large and complex photogrammetric processing tasks, is established. Experimental results with multiple kinds of remote sensing image indicate that the system is of strong capability, high automation level, great processing efficiency and convenience, and should greatly improve the remote sensing image processing capacity.