利用GPU技术使去噪方法并行化

Use of GPU technology parallelization denoising method

野外采集的地震勘探原始记录中通常包含多种噪声。噪声的存在降低了地震资料的信噪比,影响地震资料处理成果品质,因此,去噪方法研究和应用是处理工作的重要环节之一。"加权中值滤波自动检测并压制强能量干扰方法"是一种在准噶尔盆地油气勘探中非常有效的噪声压制方法。该方法在地震数据原始记录频率域剖面上采用加权中值滤波的方法自动检测可能存在的强能量干扰,并针对性地对相应频段上的噪声信号进行压制,去噪效果较为理想。但该方法的算法运行过程中涉及大量的数据计算,开发的程序需要花费大量时间才能完成一次去噪过程。提高计算效率成为该噪声压制方法推广应用的关键。高质量图像处理用途的高端图形处理器(GPU)在大规模高带宽计算方面表现出色,近年来更多地应用于高性能计算工作。CUDA并行计算开发平台帮助应用人员开发高效率计算程序,使GPU能更容易应用于高性能计算。通过分析"加权中值滤波自动检测并压制强能量干扰方法"算法实现方式,发现该算法适宜利用GPU进行并行化改造。利用CUDA并行编程技术将该算法中部分串行执行的数据计算过程改造成适合GPU计算的并行计算过程,使整个去噪方法工作效率提升3倍。GPU并行计算技术能使油气勘探数据处理过程中类似应用有效并行化,利用较小成本实现高效计算效率。

The seismic records often contain many kinds of noise. The presence of noise reduces the signal-to-noise ratio of the seismic data and influences seismic data processing quality. Noise attenuation method study is one of the important links of seismic processing. The method of Weighted median filter for automatic detection and suppression of high energy noise is a kind of effective noise suppression method applied in oil and gas exploration in Junggar basin. The method used weighted median filtering automatically detect possible high energy noise in the seismic records of original frequency domain profile and suppress the corresponding frequency band noise signal. The effect of noise attenuation is significant. But the running process of the algorithm involving large amounts of data and the program needed to spend a lot of time to complete a noise attenuation process. Improving calculation efficiency became the key in the application of noise suppression method. In large scale and high bandwidth calculation high-end graphics processor unit （GPU） shows outstanding performance and recently more frequently be used in high performance computing. CUDA parallel computing platform can be assistanted programmers develop efficient calculation program and made GPU easily applied to high performance computing applications. Through the analysis of the algorithm implementation of ＂weighted median filter for automatic detection and suppression of high energy noise＂ found this algorithm is suitable for use of GPU parallel transformation. Using of CUDA parallel programming technology adapted serial execution data calculation process of the program for GPU computing parallel computing process and which improves the noise attenuation method efficiency three times. GPU parallel computing technology can effectively parallelize the similar applications of oil and gas exploration data processing process and achieve a high computing efficiency by less investment.