Seismic random noise attenuation based on adaptive time–frequency peak filtering

• Improved intersection of confidence intervals is used to preprocess noisy signal.
• Preprocessing will finish dividing the noisy signal into signal, noise and buffer.
• Different segments will have different optimal filtering window lengths.
• Optimal filtering window lengths vary adaptively for the signal, buffer and noise.
• Compared with time–frequency peak filtering, the improve algorithm performs better.

Time–frequency peak filtering (TFPF) method uses a specific window with fixed length to recover band-limited signal in stationary random noise. However, the derivatives of signal such as seismic wavelets may change rapidly in some short time intervals. In this case, TFPF equipped with fixed window length will not provide an optimal solution. In this letter, we present an adaptive version of TFPF for seismic random noise attenuation. In our version, the improved intersection of confidence intervals 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 used in each type of segments. We test the proposed method on both synthetic and field seismic data. The experimental results illustrate that the proposed method makes the degree of amplitude preservation raise more than 10% and signal-to-noise (SNR) improve 2–4 dB compared with the original algorithm.