基于CUDA技术的离散小波变换算法研究与实现

Research and Implementation of Discrete Wavelet Transform Algorithm Based on CUDA Technology

针对离散小波变换过程比较耗时、不利于实际工程应用的问题,提出利用基于GPU平台的CUDA技术对小波变换算法做并行化改造,从而提高算法执行效率。该文分析了小波Mallat算法并行化的可行性,并详细介绍了算法的改造过程。实验表明,基于GPU/CUDA技术的并行小波Mallat算法,相较于串行小波变换算法,执行速度最高提升了50余倍,且算法效率与计算量成正向关系。

In view of the time-consuming process of discrete wavelet transform and not conducive to practical engineering applications,it is proposed to use GPU-based CUDA technology to parallelize the wavelet transform algorithm,thereby improving the efficiency of algorithm execution.This article analyzes the feasibility of the parallelization of the wavelet Mallat algorithm,and introduces the algorithm transformation process in detail.Experiments show that the parallel wavelet Mallat algorithm based on GPU/CUDA technology,compared with the serial wavelet transform algorithm,the execution speed is up to 50 times faster,and the algorithm efficiency has a positive relationship with the amount of calculation.