A Technique And Implementation Of Remote Sensing Image Real-time Processing Based On GPU

In the information age,the real-time requirement of remote sensing image processing gets more important.Besides,the resolution of remote sensing image is greatly improved with the development of radar and sensor technology,and the image data is also larger.Therefore,speeding up the image processing and managing the massive data are two problems with which remote sensing image processing in real time must be faced.GPU can accelerate image processing with its multi-core feature,which can meet real-time requirements.It is significant to research a technique that can process remote sensing images in real time with GPU to accelerate processing and database to manage image information.In this thesis,a real-time processing platform for remote sensing image is designed and implemented.This platform consists of universal hardware and has functions of processing remote sensing images in real time and managing image information by running software.Using GPU to accelerate image processing and using database to manage image information are two main features of the platform.Firstly,the hardware structure of the platform is given.The architecture of the software is designed by analyzing the demand.Secondly,the software is divided and modules of the software are realized.In the sending module and the receiving module,the stream of remote sensing images is transmitted by socket.In the storage module,information of images is stored in a relational database in which the latitude and longitude information are managed.In the memory management module,shared memory with a principle of ping-pang buffer is applied in the cross-process data transmission,which makes up the image transmission speed.In the GPU calling module,a problem is solved that too many images are overstocked seriously when the image processing speed is slower than the image receiving speed.And functions of loading images,managing information and calling GPU algorithms are realized.In the display module,the function of resizing images to fit on screen is realized.In addition,the user interfaces of the software in the platform are shown.Thirdly,the image registration algorithm embedded in the platform is also realized and tested,which is the parallelization of the image registration algorithm combining gradient and mutual information under the CUDA architecture.Finally,the function of the platform is tested.The test result shows that the platform completes the receiving and the registration for remote sensing images inreal time,and manages the information of images successfully.