Seismic records are often accompanied by impulse noise interference, which seriously affects the follow-up work of seismic data analysis. These noises are often in a concentrated state and last for a long time, so they are called impulse noise clusters. The median filter used to suppress the impulse noise clusters will greatly damage the effective signal, and the improved decision-based median filter cannot effectively distinguish the impulse noise clusters. Therefore, an impulse noise suppression method based on cluster detection is proposed. First, the short-term average/long-term average (STA/LTA) algorithm is used to distinguish and detect the noise. Then the same impulse noise cluster is distinguished to solve the problem of missing noise detection. Finally, the filter window of the adaptively stretched median filter is used to filter the detected impulse noise clusters. Synthetic and real data examples show that, compared with other methods, the method in this paper can effectively suppress the impulse noise clusters and minimize the effective signal damage.