Target Recognition And Parallel Acceleration With GPU In Marine Remote Sensing Image

With the rapid development of remote sensing technology,the acquisition of remote sensing image has shifted from the traditional synthetic aperture radar to the optical camera.The resolution of remote sensing image is getting higher and higher,which makes the detection of target on optical remote sensing image possible.As one of the most important applications of optical remote sensing images,ocean target detection has a wide application space in maritime traffic management,tracking important military targets and responding to maritime emergencies.The application of military activities is one of the most important application in remote sensing technology,many countries have invested a lot of resources in the construction of remote sensing processing platform.Remote sensing image has been widely used in several aspects of national defense military,including military target detection,battlefield situation assessment and automatic guidance of weapons and many other aspects.Whether based on the aeronautical platform or space platform,remote sensing has become the main technical means of military action battlefield information acquisition.Therefore,the research of ship target detection,automatic detection and unsupervised identification based on optical image remote sensing is of great significance in the research of naval military forces,especially the aircraft carrier formation and warship formation,which are the essential targets of investigation and monitoring.With the increasing of satellite space,the spatial resolution of image is increasing and the period of satellite re-visiting is decreasing,the data quantity of optical satellite remote sensing image is huger and huger.For the vast marine area,the ship target is sparse and small.It is still a challenging task to detect the ship target rapidly and accurately from the massive optical remote sensing image and it needs to be solved urgently.Traditional image processing algorithms are based on CPU serial implementation,because the calculation of the algorithm is complex,so the CPU is difficult to complete the algorithm for real-time calculation.In recent years,NVIDIA's unified computing architecture equipment(compute unified device architecture,CUDA)can transplant the serial code execution on the GPU to run in parallel on the CPU.The efficiency of the algorithm after transplantation compared to the CPU increases a few times,which can achieve real-time computing algorithm.This paper mainly focuses on the detection and recognition of ship targets in optical remote sensing images.This paper introduces the background of the research and the difficult problems in the detection and recognition of warship targets,and analyzes the main processing methods of ship detection and recognition based on optical remote sensing images.This paper analyzes and compares the mentioned algorithms to find out the efficiency can accelerate the optimization.This paper mainly introduces the GPU acceleration method from the aspect of algorithm execution,and analyzes its application in image processing acceleration.At last,the GPU parallel implementation of target detection task is finished.