Some factors lead to spurious signals in the multi-beam sounding process, including complex changes of the sea, the vessel motion attitede, the unreasonable setting of equipment parameters and so on. Although these factors were filtered strictly and the system errors were compensated in the post processing, residual errors still exist in multi-beam swath joins. Such errors produce uplift and abnormal phenomena in the measured terrain between swaths. Therefore, a new method of terrain spectrum analysis was established to improve the efficiency of massive data processing and to accurately reflect the submarine topography. This method was based on the blend of extraction technology of shortwave signals and fitting model of long-wave signals, among which wavelet analysis was used to extract the terrain highfrequency signals (short-wave) and multi-surface function was applied to fit the long-wave signals. Then, the two types of signals were fused to reconstruct the actual seabed topography. Finally, the indexes of depth discrepancy were used to estimate the effect of residual errors. Experimental results show that the proposed method could effectively eliminate the uplift phenomenon of the seabed Digital Elevation Model (DEM), and correctly reflect the detailed topographical variation. This work could further improve the sounding accuracy and the efficiency of swath joins, and beneficial to the fusion of multi-beam digital signals and the image information of side scan sonar.