Abstract: High-resolution seismic data can realize better well seismic calibration results,clearer structural interpretation and reservoir characterization, and also hold better identification ability of thin layers. In order to improve seismic resolution, it is necessary to carry out frequency extension processing on seismic data. The conventional seismic spectral broadening method used to be carried out in the frequency domain is susceptible to high-frequency noise and thus reduces the reliability of the data. In this paper we propose a time-domain frequency extension technology which is based on the combination of ghost wave processing and non-stationary multi-order differential resolution. The sesmic spectral can be broaded merely through multiple times of integration and differential operation, whose results are performed amplitude matching at Gaussian window to ensure the consistency of the amplitude before and after the processing. Weighted fusion is performed on the difference results, the high-order difference is assigned a smaller weight to avoid the influence from the high-frequency noise, and thus improve the anti-noise performance of the algorithm. The theoretical models and processing results of the field seismic data show that the algorithm can effectively improve the resolution of seismic data.