Abstract: When estimating the azimuth of fast-moving targets using long-term multiple snapshot data, traditional methods assume a constant azimuth for the target throughout the processing of multiple snapshots, which may lead to azimuth deviations or the emergence of false peaks, resulting in the misidentification of non-existent targets. To address this issue, this paper proposes an azimuth estimation method for fast-moving targets based on sparse Bayesian learning. Unlike conventional methods, the proposed approach does not assume a constant azimuth across multiple snapshots. Instead, it incorporates an unknown parameter to characterize the azimuth change rate. This is achieved by constructing multi-snapshot array steering vectors that more accurately model the dynamic motion state of the target. Sparse Bayesian learning is then employed to jointly estimate both the initial azimuth and the azimuth change rate for each target. Numerical simulations show that, the proposed method significantly improves the accuracy of azimuth estimation in scenarios with multiple moving targets compared to traditional sparse Bayesian learning algorithms. In addition, results derived from the processing of experimental sea data indicate that this method effectively estimates target azimuth trajectories in long-term multi-snapshot scenarios and exhibits excellent azimuth resolution capabilities.