Frequency-Aware Wavelet U-Net: A Hybrid Framework for Seismic Data High-Frequency Expansion

The seismic data are usually limited by the frequency bandwidth and lack significant high-frequency components due to attenuation in the subsurface medium and frequency filtering effects during propagation. The expansion of high-frequency components improves resolution and facilitates thin-layer identification. However, during the expansion of high-frequency components, the low-frequency information may be lost. To address these issues, we propose a novel hybrid framework: the frequency-aware wavelet U-Net (FAWU-Net) for seismic frequency expansion. FAWU-Net integrates discrete wavelet transform (DWT) and attention gate into the U-Net framework. DWT decomposes features into low- and high-frequency components: low-frequency features are preserved via skip connections to an attention gate module, while high-frequency features are enhanced and expanded through encoder-decoder processing. We have designed a frequency-spatial domain loss function and have incorporated transfer learning to further optimize the network. Experimental results confirm FAWU-Net’s capability to expand high-frequency components while retaining low-frequency features, significantly improving the resolution of stratigraphic structures.