Research On GPU Based Image Synthesis Method For GF-1 WFV Images

With the success of the GF-1 satellite launch and the routine operation,the spatial resolution of domestic satellite images has been greatly improved,and the application field of remote sensing technology has been promoted to be broader,deeper and finer.In turn,the development of application field has put forward higher requirements on remote sensing technology.At present,although the spatial resolution of remote sensing image has been improved,the actual temporal resolution of remote sensing image is not high due to the limitation of satellite technology and the influence of cloud,fog and shadow,which restricts the extensive applications of remote sensing technology.On the other hand,higher spatial resolution of satellite image means larger data volume inevitably,the efficiency of remote sensing images processing has become a problem.In order to remit the problem of lower real temporal resolution of satellite images,This paper takes GF-1 satellite Wide Field of View(WFV)image as the experimental object and takes Beijing as research eara,and builds a kind of image synthesis algorithm based on Fourier harmonic model,visual assessment and quantitative assessment were carefully apllied on the accuracy,stability and feasibility of the synthesis algorithm.And then,a software used to simulate GF-1 WFV images is developed,the software introduces GPU accelerated solution to improve the efficiency of synthesis algorithm,aiming at solve the problem of huge volume of data.Through the analysis of the results of experiments,we come to the conclusion:1.Synthesis algorithm based on the harmonic model achieves high accuracy and stability.The synthetic images and real images are very similar visually,difference of most pixels between the synthetic images and the real images is between-0.03 to 0.03,the root mean square error(RMSE)ranges from 0.02 to 0.05,standard deviation of the difference image is about 0.02.2.The accuracy and stability of the simple model,the advanced model and the complete model used in the algorithm are not positively related to the model complexity,they are determined by the complexity of the model,the number of input images and the surface coverage type.It is necessary to determine the model to be used according to the evaluation of synthetic image.3.The GPU accelerated synthesis software is easy to be operated and achieves a high executing efficiency,which achived a speedup ratio of 5 with respect to the unoptimized CPU implementation and a speedup ratio of 3 with respect to the optimized CPU implementation.