Interferometric spectral imagers enable simultaneous acquisition of two-dimensional spatial information and one-dimensional spectral data of targets. It offers significant advantages, including high throughput, multi-channel capability and superior resolution, with broad applications in agricultural vegetation analysis, soil composition assessment, marine biology studies, atmospheric monitoring and mineralogical investigations. The workflow involves dispersion, interference and imaging to capture target's images and interference information, from which spectral information is reconstructed.However, during the acquisition of interferograms, deviations arise due to factors such as the non-uniform response of the detector itself, variations in signal processing circuits across different sub-regions, environmental fluctuations affecting the detector's operation, and non-uniformity in optical energy transmission. These deviations prevent the interferograms from accurately representing the target's spatial and spectral characteristics, resulting in spectral distortion during reconstruction. Consequently, prior to spectral reconstruction, interferogram correction must be performed through steps including dark current correction, response non-uniformity compensation, and bad pixel correction to mitigate the impact of various errors on the reconstructed spectrum.The corrected interferogram represents a superposition of high-frequency information and low-frequency background noise that varies with optical path difference. The low-frequency component adversely affects the accuracy of spectral reconstruction and should be eliminated through filtering. Current interferogram filtering methodologies primarily encompass differential method, fitting method, and Empirical Mode Decomposition (EMD). While differential method inherently compromises interference curve symmetry, fitting-based method demands prior knowledge of signal characteristics and empirical mode decomposition has drawbacks such as endpoint effects and mode mixing.Variational Mode Decomposition (VMD) proposed by DRAGOMIRETSKIY K et al in 2014, is an adaptive, fully non-recursive modal variational method based on empirical mode decomposition, which has the advantages of determining the number of modes and suppressing mode mixing. Discrete Wavelet Transform (DWT) proposed by MALLAT S G in 1989, is a method that can extract local features of signals and achieve multi-resolution analysis, particularly suitable for capturing the non-stationary characteristics of signals, with advantages in multi-scale and localized analysis. To address the constraints of existing filtering methods, this paper proposes a combined VMD-DWT filtering method. Through the synergistic application of VMD and DWT, low-frequency component is systematically separated while high-frequency components are effectively extracted, achieving enhanced spectral reconstruction accuracy. First, VMD is performed on the interferogram to determine whether the correlation coefficient of each mode component exceeds the threshold, thus obtaining the optimal number of modes. Second, the interferogram undergoes VMD under the optimal number of modes, filtering out mode component with relatively high value and low center frequency to represent the low-frequency component of the interferogram. Then, DWT is used to separate the residual signals in the mode component to obtain low-frequency component. Finally, the low-frequency component is directly subtracted from the interferogram to obtain an interferogram containing only high-frequency information, completing the filtering process.Experimental validation is conducted using the HJ-2A hyperspectral imager, with the Signal-to-Noise Ratio (SNR) and the Relative Quadratic Spectral Error (RQE) of the reconstructed spectrum in the spatial dimension serving as evaluation metrics. The results demonstrate that the advantages of fitting filtering are intuitive results and fast computational speed, while the disadvantages include the need for prior information. Specifically, polynomial fitting cannot fully represent the low-frequency component, leading to significant fluctuations in the fitting curve and larger errors at both ends, which affect the filtering effect. The SNR of the restored spectrum in the spatial dimension after fitting filtering is 28.119 3, and the RQE is 0.005 268. The advantages of EMD filtering include adaptability, while the disadvantages include difficulties in determining the values of the upper and lower envelope lines at endpoints, leading to endpoint effects. The decomposed mode components cannot represent an independent oscillation mode, resulting in mode mixing. Both endpoint effects and mode mixing affect the filtering effect. The SNR of the restored spectrum in the spatial dimension after EMD filtering is 28.779 6, and the RQE is 0.004 887. The advantages of VMD filtering include the ability to determine the number of modes and suppress mode mixing, while the disadvantages include the influence of the penalty factor and the number of modes on the filtering effect. The SNR of the restored spectrum in the spatial dimension after VMD filtering is 29.054 2, and the RQE is 0.004 698. The advantages of the VMD-DWT combined filtering method include the ability to determine the number of modes, suppress mode mixing, and provide multi-scale and localized analysis. The SNR of the restored spectrum in the spatial dimension after VMD-DWT combined filtering method is 29.075 2, and the RQE is 0.004 683.The proposed VMD-DWT combined filtering method demonstrates superior performance compared to conventional approaches, showing SNR improvements of 3.40%, 1.03%, and 0.07% over fitting filtering, EMD filtering, and standalone VMD filtering, respectively. Concurrently, it achieves RQE reductions of 11.10%, 4.17%, and 0.32% for these respective methods. This innovative approach effectively eliminates low-frequency component while preserving critical spectral features, thereby improving spectral restoration accuracy and enhancing the quantitative measurement capabilities of interferometric imaging spectrometers. The combined advantages of noise suppression and signal fidelity maintenance make this methodology particularly valuable for high-precision spectral analysis applications.