<p>As hydrocarbon exploration in China progressively targets deeper and ultra-deep formations, particularly in western regions, the acquired seismic data exhibit weak reflection signals and a low signal-to-noise ratio (SNR). The intense energy attenuation of seismic waves after long-distance propagation, combined with complex geological structures and near-surface conditions, further diminishes the SNR. Consequently, the recovery of weak deep signals and suppression of strong background noise have become crucial for accurate deep structural imaging and resource characterization. This paper presents an adaptive progressive framework for weak signal enhancement and noise suppression, applied to seismic data from western China. Within a progressive denoising architecture, a bilateral filtering operator is constructed to perform robust noise estimation in both time and frequency domains, thereby reducing residual artifacts commonly observed in conventional methods. To address data heterogeneity arising from complex subsurface structures, dualdomain kernel functions are employed for noise estimation. A Laplacian mask is introduced to construct a spatially adaptive noise variance field, which accurately captures local noise characteristics. Furthermore, an adaptive control factor is incorporated to dynamically adjust the strength of the variance field, addressing severe amplitude attenuation and noise uncertainty in deep layers while preserving weak signal integrity. The proposed method has been applied to both synthetic data and field data from a western China exploration area. The results indicate that this method effectively suppresses strong noise while maximally preserving weak signals, leading to an improved recovery of deep, weak reflections.</p>