Parallelization on Model of Ecological Environment Remote Sensing Evaluation Based on GPU

LI Lin,GU Jinfeng,SONG Anjie,ZHENG Haining,CAO Jin,ZHU Dehai

DOI: https://doi.org/10.6041/j.issn.1000-1298.2017.05.016

2017-01-01

Abstract:In order to solve the problem of the model's slowly processing speed of ecological environment remote sensing evaluation currently,a framework about GPU image processing was designed with data partitioning and scheduling asynchronous transmission which was based on the in-depth analysis about the data transmission bottleneck of heterogeneous CPU + GPU general computing platform.It included the carbon fixed quantity and grassland degradation index,whose intrinsic parallelism met the GPU computing features.For the above models,it was put forward based on CUDA parallel implementation.The core link of indexes for evaluation of ecological environment of remote sensing data standardization and weighted fusion of CUDA parallel module were implemented.Finally,the effectiveness of technical methods was verified through experiments,as the scale of data became larger,the parallel execution speed of three business models became faster,the speedup ratio of the fixed amount of carbon achieved a 8.04 times execution rate lift;the speedup ratio of the index of grassland:degradation achieved a 12.21times execution rate lift;and the speedup ratio of the index of ecological environment achieved a 7.45times execution rate lift.At the same time,the speedup ratio was decreased as the number of input data files increased,equipment between I/O was still the main factor which restricted the running efficiency of the algorithm.