Frequency estimation of sine wave signals is a common problem in the radar field. When the true frequency approaches the quantization frequency points， the calculation of the frequency shift factor in the I-Rife algorithm can introduce significant errors. In order to improve the accuracy of frequency estimation， this paper analyzes the performance and error sources of the Rife and I-Rife algorithms. By utilizing a spectral refinement method， a modified I-Rife algorithm is proposed. It replaces the amplitude of the spectral peak point with the amplitudes at 0.5 points to the left and right of the peak point， and interpolates the amplitude using the second highest frequency point. This approach allows for a more accurate estimation of the frequency offset. The proposed algorithm effectively enhances the estimation accuracy of frequency while maintaining a similar computational complexity to the original I-Rife algorithm. Simulation results demonstrate that the improved I-Rife algorithm outperforms the original I-Rife algorithm in overall performance and achieves an estimated root mean square error closer to the Cramér-Rao lower bound.