Adaptive Time-frequency Peak Filtering For Seismic Random Noise Attenuation

The seismic exploration is an important way to explore oil and gas resources, which is based on the seismic waves motivated by manpower, and applies seismic sensors to collect the reflection signal, in order to solve the problem of structural geology. That is to say, seismic exploration is used to find oil and gas resources. However, the reflection signal is always corrupted by a variety of noise, which seriously affects the valid information acquisition. Particularly for recent, due to the over-use, the oil and gas in shallow surface is not enough to cope with demand. So the exploration develops into deeper layer. But the deeper we explore, the weaker the valid signal is, and the stronger the random noise is. Therefore, seismic data denoising is top priority.The time-frequency peak filtering(TFPF) is a time-frequency-based algorithm. Recently, TFPF has been widely applied to seismic random noise attenuation. Whereas, TFPF still has some shortcomings. 1. The conventional TFPF uses a fixed window length for all frequency components. As a consequence, serious loss of the valid information or insufficient suppression of the noise is unavoidable. 2. TFPF carries out filtering only along the time direction, which lacks consideration of the spatial correlation. 3. TFPF is approximately equivalent to a time-invariant low-pass filer, which means that the frequency components of signal higher than some cut-off frequency would be attenuated.In our paper, we demonstrate that TFPF is really approximately equivalent to a time-invariant low-pass filer through the essence of TFPF. Consideration of question 1, time domain adaptive TFPF(T-ATFPF) is proposed. In this version, the intersection of confidence intervals based on Chebyshev inequality combined with short-time energy criterion is used to preprocess the noisy signal. And then, we choose an appropriate threshold to divide the noisy signal into signal, buffer and noise. Different optimal window lengths are applied in each type of segments. We test the proposed method on both synthetic and field seismic data. The experimental results illustrate that T-ATFPF makes amplitude preservation and noise attenuation better than the conventional TFPF.Although T-ATFPF obtains good filtering effects, it still does not solve question 2 and 3. Meanwhile, T-ATFPF is not suitable for the situation with the low SNR. For the purpose of solving these problems above, we propose the spatiotemporal adaptive TFPF(ST-ATFPF). In this approach, the noisy signal is first transformed into Radon domain by high-resolution Radon transform, which contains temporal and spatial information. Subsequently, we construct the convex hull of the filter set, which is composed of the impulse responses of TFPF with different window lengths, and minimize a convex functional to find the optimal path by Viterbi algorithm. Finally, we get the spatiotemporal filtered results by inverse high-resolution Radon transform.In order to test the effectiveness of ST-ATFPF, the same synthetic and field seismic data is filtered by the conventional TFPF, Radon domain TFPF(R-TFPF), T-ATFPF and ST-TFPF, respectively. The experimental results demonstrate that the validity of ST-ATFPF with higher output SNR and less signal information loss than the other three methods.