Design And Implementation Of JPEG2000 Remote Sensing Image Decompression System Based On CPU/GPU

Although JPEG2000 is a well-developed image compression standard,it is not widely used in applications because of its high computational complexity.Currently,it is mainly used in the compression of remote sensing images.The JPEG2000 image compression standard is used by most remote sensing images in China's Gaofen satellite now.For the purpose of real-time compression,multiple chips are used to compress images in parallel on the satellite.For the purpose of real-time decompression,cluster devices are used to decode images in parallel in the ground processing system.The huge cluster devices of the ground system severely restrict the flexible application of remote sensing images in small application units and mobile facilities.In order to solve the problem of miniaturization of the existing ground decompression system,a real-time JPEG2000 remote sensing image decompression system is designed and implemented in this thesis based on the CPU and GPU platform.The main works done in this dissertation are as follows:First,in order to reduce the number of cluster devices in the ground remote sensing image decompression system,multi-threading technology is used in this thesis.Based on this problem,this thesis uses multi-thread technology to realize the decompression of remote sensing images on a multi-core CPUs computing platform.This parallel processing method can reduce the number of cluster devices and achieve the purpose of real-time decompression.In this regard,this thesis analyzes the characteristics of the remote sensing image decompression system and the multi-core CPUs computing platform.A lightweight remote sensing image decompression system is designed and implemented based on CPU.Secondly,multi-threaded parallel decompression system can be implemented with multi-core CPUs,but multiple computers are still needed to achieve real-time decompression.Therefore,lightweight system is not truly miniaturized.Based on this problem,GPU accelerators are loaded into a computer with multi-core CPUs for decompression processing.The GPU can handle huge amounts of data in parallel by a large number of cores with smaller frequency.With small GPUs on one computer can achieve real-time decompression of remote sensing images.In this regard,this thesis analyzes the characteristics of each decompression module and GPU computing platform.A miniaturized remote sensing image decompression system is designed and implemented based on GPU.Finally,this thesis implements functional testing and performance testing of a miniaturized remote sensing image decompression system.Decompress the compressed stream with the realized system to verify whether the reconstructed image can be output correctly and test whether the decompression speed is not lower than 220 Mbps.