Design And Implementation Of The MAP Tile Creation System Based On GPU

With the continuous integration of computer technology and geographic information,the applications of Web-based geographic information system are becoming more widespread.And they are no longer limited to traditional industries such as agriculture and meteorology.Nowadays,in order to speed up the display of map data on the client,the tile pyramid structure is mainly used to store and manage image data in Web GIS.With the continuous advancement of aerospace science and remote sensing image technology,massive raster map data is gradually accumulating.How to construct tile pyramids of image data quickly is an urgent problem for high-performance Web GIS.The traditional tile pyramid construction methods are mainly divided into serial single-machine and distributed cluster.However,GPU provides a new platform for slicing massive map data because of its powerful parallel processing capability.This thesis uses CUDA parallel computing architecture to design a fast slicing method for raster map data.The data re-sampling and the parallel operation of tile JPEG compression are all completed in GPU during tile pyramid construction,which changes the traditional way of generating the tile pyramid.The main research content of this thesis is as follows:(1)Several parallel algorithms commonly used for image scaling in GPU are studied and applied to tile scaling under different map projections.(2)JPEG compression standard is researched.Through taking advantage of the GPU-CPU heterogeneous computing performance fully,the JPEG compression of tile data is achieved.The parallel parts of JPEG compression are completed on the GPU side such as the color transformation and sampling,DCT transformation,quantization.The entropy coding part with higher logic operation is completed on the CPU side.(3)According to the characteristics of the tile pyramid,the implementation of the GPU-based quadtree index map slicing algorithm is deeply analyzed.In order to improve the utilization of GPU memory,a stripe-level slicing algorithm based on GPU platform is proposed,which generates tiles of the current zoom level by iteratively processing the divided map strips.It uses cache technology to generate border tiles and high-level tiles of pyramids,and reduces data transfer between disk and host memory as well as host memory and GPU.(4)The system of map slicing based on GPU is designed and implemented by taking advantage of modularization.The metadata module provides the meta information required for the slicing map Web service distribution.The space management module guarantees the stability of the slicing system and meets the need for frequent requests of same size memory by using a multi-level linked list allocation pool.Tiles storage module designs a unified interface to support different storage systems.Based on the above work and research,map tile creation system in GPU is designed and implemented.Through publishing the WMTS service,the correctness of tile pyramids generated by the slicing system is verified.Performance tests show that using GPU to perform map tile creation has performance advantages.Compared with CPU,using GPU can significantly improve the slicing speed of massive image map data.