Acceleration-domain seismic data acquired by digital nodal instruments, when converted to the velocity domain through time integration, can effectively enhance low-frequency signals and thereby benefit deep imaging. However, this process also nonlinearly amplifies unstable components such as surface waves, external interferences, and low-frequency random noise, leading to a reduction in low-frequency Signal-to-Noise Ratio (SNR), limitations in velocity model building, and further masking of deep reflection energy. Therefore, how to suppress the low-frequency noise amplified during integration while preserving effective low-frequency signal energy has become a key technical challenge in the processing of digital nodal seismic data. To address this issue, this study proposes a multi-feature adaptive low-frequency noise suppression technique. According to the generation mechanisms and characteristics of different types of noise, a dispersion-curve-based surface-wave suppression method is employed to accurately separate and attenuate surface-wave interference by extracting and analyzing surface-wave dispersion characteristics. In addition, an external interference suppression method based on MAD-Trimmed vector median filtering is introduced, and a dynamic-statistics-based threshold optimization technique is incorporated to suppress low-frequency background noise. The processing results demonstrate that the proposed method can effectively suppress complex low-frequency noise while preserving the integrity of effective low-frequency signals. Compared with conventional denoising approaches, the proposed technique achieves a better balance between noise attenuation and signal fidelity, improving the signal-to-noise ratio of velocity-domain data by approximately 1.3 times, and providing more reliable, high-quality input data for subsequent seismic data processing.