To obtain high-quality and effective seismic data, it is necessary to remove the random noise associated with the exploration process in actual seismic exploration. The random noise in seismic exploration in desert areas has the characteristics of low frequency, nonlinearity, non-stationarity, non-Gaussian, and effective signal and noise spectral overlap. A method combined with unsupervised feature learning and TFT ( Time- Frequency Transform) technique is proposed to reduce random broadband noise in desert seismic data. VMD (Variational Mode Decomposition) is an effective time-frequency decomposition method. Using its excellent time-frequency decomposition characteristics, the desert seismic signal is decomposed into several modes with different components. The SF(Sparse Filtering) algorithm is used to identify the effective signals of each modal component, achieving the separation of signals and noise. Both simulation and field experiments confirm that the proposed method achieves effective suppression of random noise while maintaining the fidelity of useful seismic signals, offering a robust technical basis for acquiring high-quality seismic data in desert environments.

variational mode decomposition ( VMD), desert seismic random broadband noise, denoising, feature learning, sparse filtering