Abstract: Accurate ship velocity inversion is fundamental to maritime traffic management and situational awareness. Conventional SAR modes such as ScanSAR generally provide short synthetic aperture times and relatively low spatial resolution, limiting high-accuracy velocity inversion for moving targets. By contrast, sliding spotlight SAR uses a longer synthetic aperture time and thus retains richer target-motion information, making it promising for ship velocity monitoring. However, a ship's velocity vector may vary appreciably during the long integration interval, violating the constant-velocity assumption used in conventional inversion methods. To address this limitation, we propose a two-dimensional ship velocity inversion method that combines sublook decomposition with motion compensation. The method first models each ship as a variable-velocity moving target. Sublook decomposition then imposes physically based motion constraints on the phase-refocusing search space, enabling accurate estimation of the azimuth velocity. The method also accounts for heading deviation and combines a high-accuracy heading estimation algorithm with vector decomposition to retrieve the two-dimensional ship velocity. Experiments with measured GF-3 sliding spotlight SAR data yielded a mean absolute error (MAE) of 0.25 m/s and a mean absolute percentage error (MAPE) of 7.39%. The method also performed well in estimating the azimuth velocities of anchored ships, indicating its potential for monitoring slow-moving maritime targets.