The work aims to propose a multi-threshold singular value decomposition (MTSVD) method based on singular value decomposition (SVD) and Hanke matrix to reduce the noise of source data, make it closer to the theoretical value and reduce the effect of experiment error on subsequent calculation. First, the obtained frequency-response function (FRF) of the system was added with 10% random error. Then, the MTSVD method proposed in the paper was used to reduce the noise. Finally, the results were compared with the prediction results without noise reduction and the noise reduction results obtained by singular value accumulation. After the MTSVD method was used to process the frequency response function of the coupling system obtained from the experiment, the modified value was closer to the theoretical value, and the noise reduction effect of this method was better than the existing singular value accumulation method. The verification results show that the MTSVD method proposed in the paper can effectively reduce the noise, decrease the error carried by the test measurement source data and make it closer to the theoretical value. Therefore, this method has prominent feasibility and application prospect in transport packaging.