Due to complex acquisition environments and geological conditions, in desert seismic data, the background noise is always intense and overlaps with the frequency range of the seismic signals. This severely blurs the seismic signals, bringing challenges to accurately extract desired reflection information. Effectively suppressing random noise and significantly optimizing the signal-to-noise ratio (SNR) is emerging as a key issue in seismic data processing. A Multi-scale Feature Interaction Enhancement Network (MFIEN) for intense background seismic noise attenuation in desert areas is proposed in this paper to address this problem. In general, MFIEN has a multi-scale feature interaction structure that combines feature information from different layers and captures distinct features through effective information integration. Furthermore, a fusion feature enhancement module (FFEM), incorporating dilated convolutions and convolutions with different kernel sizes, is proposed. This expands the receptive field without changing the size of the feature maps, thereby preserving the structural features of seismic records more effectively. Both synthetic and field desert data denoising results indicate that MFIEN can accurately suppress intense background noise and effectively recover weak signals, significantly enhancing the quality of seismic data.
Read full abstract