摘要
深层勘探及小断裂、小幅度构造等隐蔽油气藏勘探是目前老油区增产的重要手段,而进行这类油藏勘探时所面临的共同问题是在相对较高的噪音中识别信号,即弱信号的识别问题.利用小波的多尺度特性,可以在一定程度上压制噪声,识别弱信号.但小波变换在处理二维数据时,对线性奇异性的边缘的识别有一定局限性,而曲波变换能较好地识别二维奇异性边缘.基于曲波变换的去噪方法大多数都建立在曲波变换能最稀疏地表示具有线性边缘图像这一基础上,去噪过程也就是使曲波系数稀疏化的过程,一般是阈值法或其变通.但对于弱信号来说,曲波系数并不是越稀疏越好,当噪声的强度大于弱同相轴的1/2时,由于信号所对应的小值曲波系数被衰减,阈值法会造成有效信号的断点和形态的失真,也就是说去噪与弱信号识别是一对矛盾.本文从具有较高去噪能力的曲波阈值去噪出发,探讨了曲波变换去噪对弱信号的影响,并根据地震资料弱信号的特点,提出降低阈值并采用线性压制补偿小系数的方法来保护曲波系数值较小的弱信号的特征,同时采用曲波域平滑的方法来保证去噪效果,从一定程度上缓和了这一矛盾.通过实验,可以看出这种方法可以从最大幅值为2倍弱信号强度的噪声中识别出弱信号,并且信噪比能得到较大提高.
It is very important for old exploitative oil regions to prospect deeper layers and low-amplitude paleostructures. Inevitably, we must face the problem of detecting weak signals mixed with strong noise because of the complexity of those geologic objects in exploration. With multi-scale characteristic, curvelet transform could remove most of the noise and detect weak signals. Denoising methods in eurvelet domain are based on the principle that the curvelet transform has the optimal sparse representation for image edges. Those methods such as thresholding try to obtain a sparser curvelet coefficient matrix which stands for signals and they can get remarkable effect. But when we put a high threshold in curvelet-denoising to suppress strong random disturbance, the signals estimated will have dim breakpoints and anamorphic events because a part of small curvelet coefficients for valid signal are removed. To protect weak wavefront, we introduce a process which can decrease the threshold and retains some small effective coefficients. Firstly, we transform a noising image to eurvlet domain and then suppress the small coefficients by thresholding which has a low threshold value. Because of the lower threshold value, many noise coefficients are remained as wild values in curvelet domain. In the second step we clear wild value among coefficients away using median filter and curvlet domain is smoothed. At last we transform the coefficient matrix to time-space domain and obtain a denoised image. The reconstituted image we got has a higher SNR and a fewer loss of weak information compared to former thresholding when it was put to model and real seismic data. We find that this program can remove most noise and protect weak signals when the threshold value selected as half of that of hard thresholding. Experiments prove that the denoising method based on curvelet transform can be combined with other techniques and it may give a better result.
出处
《地球物理学进展》
CSCD
北大核心
2011年第4期1415-1422,共8页
Progress in Geophysics
关键词
曲波变换
地震弱信号
阈值法
曲波域平滑
curvelet transform, seismic weak signal, thresholding method, smoothing in curvelet domain
