The Research On Key Technology Of Aviation Remote Sensing Images Cooperative Storage And Processing Under Heterogeneous Environment

Aerial images have been widely used in urban planning, land resources management, mapping, agriculture, forestry, traffic, environmental protection and other fields, and provide a strong protection for the economic development of society, the interests of the state and defense security. But as the expansion on scope of application of GIS technology and space information service grew in popularity, how to obtain the spatial information products by storage and processing the collected aviation remote sensing image quickly put forward higher request to the processing mode and speed. This article mainly disusses the distributed storage data scattered, based on the remote sensing image processing GPU, heterogeneous cluster CPU/GPU coordination treatment under such key technologies by analysising and researching some main problems such as remote sensing image data efficient, safe storage and fast calculation, and provides technical suppor for calculating the development with independent intellectual property rights high-performance aviation remote sensing data fast storage and automatic processing system. This paper main research work includes:(1) The paper discusses the key technology of vector data spatial conflict detection based on GPGPU. It firstly describes the details of the GPGPU development present situation, the principle, application methods, and the realization of GPGPU computing model based on CUDA (Compute Unified computing devices Architecture), and analyses the necessity and basic principle of remote sensing image mapping extraction vector data spatial conflict route, and detects the spatial conflict between river line and contour line as an example, describes the detailed how to use CUDA realize parallel for collision detection method based on topological space relationship and present mission packet processing mode and dual precision conversion method for fitting any hardware conditions and different mission packet size. Finally experiments in different mission data size and result shows the great advantage of GPU parallel processing and feasibility of the GPU spatial conflict dectect processing.(2) Based on analysis of the aerial images distributed storage analysis, this article discusses the present research achievement and problems between the two mainstream data storage strategy:redundancy copy and secret sharing, and showed the system availability and reliability, safety and performance of IDA scattered algorithm from secret sharing methods by compared with redundancy copy method thoroughly, and confirmed effectiveness the IDA data separation algorithm which is applied to aviation remote sensing image distributed storage, and proposed the fast GPU oversized thread parallel IDA coding algorithm for low efficiency original IDA data coding algorithm, which combined partitioned matrix methods, could realize the sharing memory fast data access by caching different file blocks bytes, experimental results prove the file size of dedicated I6M blocks of IDA coding has can achieve real-time or quasi real-time requirement, and can meet the reliable, efficient storage requirements of aerial images.(3) More and more large amount of aerial image data has been unable to achieve rapid production practice demands solely rely on photographing measurement of theoretical innovation, algorithm improved and model optimization method. This paper analyzes the theory of parallel algorithm, and discussed the parallelism of aerial image structure, including parallel mode, the load balancing based on GPU computing, and the process of aerial image processing, and based on image processing basic process include image matching and image pre-processing, splicing, and separately implemented by histogram equalization, template matching and phase correlative stitching parallel algorithm based on the GPU. With the experiments show that GPU parallel computation is superior to the CPU computation and it laid the foundation for realizing other core steps parallel algorithms of remote sensing image processing.(4) Present the researcher of parallel processing often only independently studied CPU or GPU, less discussion between the CPU and GPU, and with the present situation of limited collaborative computing longitudinal extension ability we cannot meet fast storage and processing demand problems with a large amount of aerial images. Based on parallel computing and parallel computing cluster theory, this paper puts forward the heterogeneous cluster which builted on MOCG CPU/GPU collaborative parallel computing framework, the framework realizing the heterogeneous parallel computing cluster computing resources to maximize utilization through the MPI, OpenMP, CPU/GPU three-layer parallel mechanism, and using the sender/receiver launched task scheduling mechanism according to right value table realization. Meanwhile we use the IDA data scattered algorithm and aerial images histogram enhance for experiment and analysis of heterogeneous cluster parallel processing efficiency and problems in different data input environment, different heterogeneous cluster scale, and under the condition of different data quantity size comprehensive experiments for different conditions. Experimental results show that the MOCG CPU/GPU heterogeneous cluster parallel computation framework is better achieve task scheduling and computing resource utilization, meet large data quantity aerial image parallel storage and processing requirements, and provide practical basis and beneficial reference for mass aerial images distributed storage and management in the further.