Converting acceleration-domain seismic data into velocity-domain seismic data using numerical integration introduces integration noise, which adversely impacts subsequent seismic data processing. This paper initially analyzes the reasons for integration noise in both time-domain and frequency-domain integration. It then conducts comparative analyses of various methods for suppressing integration noise, including high-pass filtering, polynomial fitting, and EMD-based methods, using real seismic data. Additionally, the paper examines the practical applicability of these methods and briefly discusses the feasibility and cost-effectiveness of using time-domain integration combined with EMD for conversion of acceleration-domain seismic data into velocity-domain seismic data in on-site processing. This technology has been successfully applied in the X project in South America. During processing, acceleration-domain seismic data collected by digital nodes were transformed into velocity-domain seismic data and merged with velocity-domain seismic data received by analog nodes, resulting in improved data consistency and an enhanced signal-to-noise ratio of the weak seismic signal from deep, thus confirming the method's effectiveness.