The Implementation Of Miniature UAV-borne SAR Real-time Imaging Software Based On GPU

In recent years,airborne miniature unmanned aerial vehicle(UAV)SAR is increasingly used in military field,such as battlefield reconnaissance,target acquisition,accurate positioning,etc.With the development of technology,the digital processing technology for SAR system on real-time imaging has become the mainstream of the research.The high resolution of UAV-borne SAR makes the calculated amount and data throughput of real-time processing bigger and bigger,and in order to ensure long-time endurance of UAV,the choice of low power consumption processing system is becoming more and more important.Traditional processors have been difficult to meet the demand of low-power and real-time imaging of UAV.In this paper,the miniature UAV-borne SAR real-time imaging software is designed and implemented in parallel processing method.Firstly the principle and the application of parallel computing based on GPU are explained,and according to practical engineering application,the basic characteristic of the GPU computing platform and the program structure are introduced.And the different implementation method of high resolution airborne SAR technology is analyzed,the characteristics of the current international mainstream processors like CPU,DSP,FPGA,PowerPC and GPU are compared and the low-power mobile GPU TX1 is chosen as the airborne real-time processing platform.Based on analysising calculation process and computational complexity of the algorithm,the phase error estimation of parallel processing key technology is designed and implamented.This key technology compares the extremum of image data in parallel,and brings a new data structure in for parallel research and locating the elements.It implementes the acceleration of the algorithm from serial to parallel processing and optimizes the iterative estimates global memory access and the calculation form of phase gradient estimation.In addition,the angle correction parallel processing technology solution is designed and implemented,which increases the degree of parallelism by dynamically launching new threads in each thread.It allows parent thread grid and child thread grid to collaboratively process,which classifies more fine-grained tasks.At last,the flow of the miniature UAV-borne SAR imaging software system is designed.This paper describes the design idea of software hierarchy and the basic mechanism of kernel framework the software using.Combined with data parallel computing on GPU,tasks level parallel processing on CPU is designed,and this paper introduces the multithreaded task division and the data flow between threads.Through experimental data we validate the design of software system and shows the computation time and results of the different dimension of the experimental images,which means on the premise of meeting the demand of low power consumption of UAV,this software system correctly completes tasks for real-time processing of high resolution SAR images...