基于CUDA技术与GCV准则的小波变换压制地震面波的研究  

Wavelet transform based on GCV criterion and CUDA technology for ground roll attenuation

作　　者：许明炬 高金耀[1] 胡昊 周勐佳 XU Ming-ju;GAO Jin-yao;HU Hao;ZHOU Meng-jia(The Second Institute of Oceanography, State Oceanic Administration, Key Laboratory of Submarine Geosciences, Hangzhou 310012, China;School of Earth Sciences, Zhejiang University, Hangzhou 310027, China;East Sea Marine Environmental Investigating and Surveying Center, State Oceanic Administration, Shanghai 200137, China)

机构地区：[1]国家海洋局第二海洋研究所,海底科学重点实验室,杭州310012 [2]浙江大学地球科学学院,杭州310027 [3]国家海洋局,东海海洋环境调查勘察中心,上海200137

出　　处：《地球物理学进展》2018年第2期760-768,共9页Progress in Geophysics

基　　金：科技部重大仪器专项“海洋--航空重力仪研制”(2011YQ12004505); 全球变化与海气相互作用(GASI-GEOGE-05)共同资助

摘　　要：压制噪声的同时能够最大程度的保留有效信号是滤波处理过程中的难点.在对地震记录频谱进行分析的基础上,认为基于GCV准则的离散小波变换自适应阈值萎缩方法可以有效分离地震记录在低频区的有效信号和面波噪声.通过该方法滤波处理,有效压制了地震记录的面波噪声,使得地震记录的信噪比得到大幅度提高.且自适应阈值处理降低了人工数据处理对地震记录造成的主观偏差,使得处理结果更为精确可靠.在前述方法研究的基础上,进一步分析讨论了地震信号小波变换滤波算法的GPU/CPU协同计算模式,并将其应用于实际地震记录的滤波处理.通过对比单独使用CPU计算和GPU/CPU协同计算模式的时间花销,后者计算效率得到显著提升,整体加速比最高达到了22倍.The key of signal filtering is to preserve effective signals to the maximum degree while suppressing noise. According to the seismic spectrum analysis,a valid adaptive threshold discrete wavelet transform based on the General Crossing Validation（ GCV）criterion method is presented,which can effectively separate the ground roll signal from the reflection signal in the low frequency domain. After this wavelet transform filtering,the Signal to Noise ratio（ SN） of seismic record is significantly increased. Moreover,the adaptive threshold processing reduces the subjective deviation caused by the artificial seismic data processing,and makes the processing results more accurate and reliable. Based on the study of the above processing,the GPU/CPU collaborative computing methods of seismic signal wavelet transform filtering algorithm is further discussed and analysed,as well as been applied to the real data. By comparing time consumption,the cooperative processing of both CPU and GPU is much more efficient than the central processing of CPU,as the whole speedup ratio has peaked at 22 times.

关 键 词：GPU计算 离散小波变换 面波衰减 GCV准则 自适应阈值 

分 类 号：P631[天文地球—地质矿产勘探]