The GPU Algorithm Of Remote Sensing Image Classification And The Design And Implementation Of Web Parallel Processing System

As a research hotspot of remote sensing image processing,image classification is closely related to most remote sensing applications.Therefore,the study of fast speed has direct influence on the development of remote sensing image processing applications.With the ever-increasing scale of remote sensing data,traditional serial computing has been unable to meet the demands of computing complex remote sensing classification and real-time application of users.The application of parallel computing,such as GPU general calculation and MPI parallel,opens up new ideas for improving image processing speed.Since its birth in the mid-1960 s,computer network technology has been developing in all fields.How will the computer network technology and the combination of remote sensing image processing system,the management of remote sensing image effectively,quickly,to blend in the wave of Internet development,become a new topic for today's remote sensing field.Based on remote sensing image classification is one of the more common in dealing with the three algorithms,namely the KNN,nearest neighbor algorithm SNN algorithm and conduct the thorough research to the simple bayesian image classification algorithm,is proposed for remote sensing image classification based on GPU parallel algorithms and optimization strategy,based on a variety of specific image processing algorithm,in view of the application in image processing in fact,we design and implement of the remote sensing image processing,parallel computing system.The main work and contribution of this article are reflected in the following aspects:(1)in-depth understanding of cpu-gpu heterogeneous computation and MPI parallel computing mode.Master the GPU architecture,CUDA programming model and MPI programming model to provide theoretical basis and technical guidance for the parallel design of algorithms.(2)design and implement the parallel algorithm of remote sensing image classification based on GPU.With KNN nearest neighbor algorithm SNN algorithm and simple bayesian image classification algorithm as the research object,analysis program can be parallel hot spots,and the structure of the data,memory optimization,the optimized thread's.Through the experimental analysis,the GPU parallel programs of the three algorithms obtained an average 72.472,149.536,125.39 times speedup effect.(3)design and implement the parallel processing system of remote sensing image based on Web.The system adopts B/S model,based on ASP.net and database technology development,finally realizes the integration of three classification algorithm(KNN,SNN,simple bayesian)and two types of parallel platform(CUDA and MPI)web remote sensing parallel processing system.This paper describes the operationprocess,structure design and function module of the system.