Research On Acceleration Method Of Remote Sensing Image Processing Based On GPU Asynchronous Stream

The volume of high-resolution remote sensing data has exploded with the advances of satellite remote sensing technology,and remote sensing data has shown the characteristics of big data such as broader perspective,polyphyly,authenticity,real-time and magnanimity gradually.Moreover,the real-time demand of remote sensing image processing in various application fields is also getting higher and higher.The infinite demand of application promotes remote sensing image processing to employ multi-core parallel,cluster parallel and heterogeneous parallel computing.Relying on the advantages of the hardware architecture,the CPU+GPU heterogeneous acceleration platform is favored in the field of remote sensing image processing,and has achieved effectively improvement in performance.However,there are still problems to be solved when using the GPU to accelerate remote sensing image processing,including whether the asynchronous stream of GPU can further improve the acceleration performance and how to deal with the redundant consumption of the asynchronous streams of GPU in practical applications.In response to these problems above,the research works of this paper are as following:1.The efficiency of GPU asynchronous stream in accelerating remote sensing image processing is verified,and a Normalized Difference Vegetation Index(NDVI)extraction method based on GPU multi-stream concurrent parallel model is proposed.Theoretical analysis of the acceleration mode and execution characteristics of the two typical parallel models,GPU multi-thread parallel model and GPU multi-stream concurrent parallel model,were performed,and parallel NDVI extraction methods were implemented based on the two GPU parallel models,respectively,and compared at the same time.The experimental result shows that the GPU asynchronous stream is more efficient in accelerating remote sensing image processing and has good scalability.2.Aiming at the redundant consumption of GPU asynchronous stream in accelerating remote sensing image processing,a storage structure optimization method was proposed.By analyzing the execution characteristics of the GPU asynchronous stream,it can be seen that merging scattered GPU asynchronous stream tasks can effectively reduce the fixed cost of redundancy and the consumption for calling gaps in the GPU asynchronous stream executive process.Based on the above analysises,a method which called combined storage structure for the storage of remote sensing images is proposed to improve the performance of GPU asynchronous stream.The method proposed can effectively reduce the redundant time consuming during the execution process of multiple asynchronous streams of GPU.3.Aiming at the problem of how to use GPU asynchronous stream effectively in accelerating remote sensing image processing,a dynamic adaptive acceleration method of remote sensing image processing based on GPU asynchronous stream is proposed based on the research results above.The method proposed can dynamically obtain dynamic acceleration parameters that are suitable for load balancing operations between asynchronous stream of GPU and computing tasks according to the GPU device hardware parameters and information of the remote sensing image data to be processed.Based on the dynamic acceleration parameters,the proposed dynamic adaptive acceleration method can perform load balancing on remote sensing image processing tasks dynamically,simplify GPU asynchronous streams configuration and take fully advantage of the GPU asynchronous streams.The efficiency and good scalability of the dynamic adaptive acceleration method was proved by the experimental results.Theoretical analysis and experimental results show that the GPU asynchronous stream has a good prospect of application in the field of accelerated processing of remote sensing images.Effective optimization strategies for several performance bottlenecks of GPU asynchronous stream in accelerating remote sensing image processing,and the two accelerated methods of remote sensing image processing based on GPU asynchronous stream were proposed in this paper.The work has certain reference value for the application and performance optimization of GPU asynchronous stream in the field of remote sensing image processing,and provides corresponding theoretical and data support for the subsequent research work.