IntroductionDuring seismic data acquisition, noise interference and bandwidth limitations often degrade signal-to-noise ratio and resolution, thereby compromising geological interpretation.MethodsTo address this issue, we propose an adaptive dual-attention wavelet denoising and enhancement network (AdaWaveNet) based on a trainable wavelet feature extractor. The network is built upon trainable discrete wavelet transform (DWT) and inverse DWT (IDWT) kernels, which are constrained to satisfy the perfect reconstruction (PR) property. An adaptive thresholding function is designed to selectively suppress noise while preserving effective signals. The wavelet feature extractor enables multi-scale sub-band feature learning, and attention mechanisms are incorporated to focus on critical geological features.ResultsExperiments on synthetic post-stack time-domain data, the New Zealand Kerry3D dataset, and field seismic data from a practical survey demonstrate that AdaWaveNet consistently outperforms conventional wavelet thresholding, DnCNN-SDC, SeisGAN, and other mainstream methods in terms of peak signal-to-noise ratio (PSNR), structural similarity index measure (SSIM), edge preservation index (EPI), and high-frequency energy ratio (HFR).DiscussionThe proposed method effectively restores event continuity and thin-bed reflection information, thus providing high-quality data support for subsequent seismic interpretation.