Abstract: The time-varying non-uniform rotation component caused by the maneuvering of non-cooperative targets can lead to the non-linearization of the Migration of Range Cell (MTRC) and high-order phase terms, resulting in severe defocusing in the final Inverse Synthetic Aperture Radar (ISAR) image. This paper focuses on the core issue of compensating for the time-varying non-uniform rotation component and reviews the relevant research progress from the technical framework perspective: First, conduct a comparative analysis of the adaptability of two compensation methods: non-parametric time-frequency distribution imaging and parametric mathematical model estimation, to reveal the performance boundaries in non-uniform rotation scenarios. Secondly, based on the experimental performance of existing typical algorithms in strong maneuvering target tracking and low signal-to-noise ratio imaging, summarize the theoretical limitations of traditional methods in dimensions such as non-stationary motion modeling and low signal-to-noise ratio phase estimation. Finally, from the two technical paths of end-to-end reconstruction network in the image domain and motion parameter estimation network, analyze the current development status of the hybrid compensation architecture that integrates physical model priors with deep learning, providing theoretical references and development directions for improving the ISAR imaging quality of complex moving targets.