Abstract: Channel wave seismic exploration is an important method for fine detection of hidden small structures inside the coal mining face. Extraction of channel wave signal from seismic data is the most key step in channel wave seismic exploration. However, the channel wave is a kind of dispersion wave. At present, it is a difficulty to extract high-precision amplitude preservation channel wave signal from the complex seismic signals in which various wave patterns interfere with each other and overlap. S-transform can analyze signals in time domain and frequency domain at the same time, and has advantages in time-frequency characteristic analysis of non-stationary dispersion channel wave signal. However, the Gaussian window function of the traditional discrete S-transform algorithm loses symmetry. As a result, the negative frequency part of the time-frequency spectrum of the signal is distorted after the traditional discrete S-transform. When the traditional discrete S-transform is used to seismic data filtering and channel wave signal extracting in time-frequency domain from the channel wave seismic data, due to the distortion of time-frequency spectrum in the negative frequency part, the energy distribution in the time-frequency spectrum is inconsistent with the actual time-frequency distribution of seismic signal energy, so time-frequency spectrum of negative frequency cannot participate in filtering and inverse transformation, resulting in the amplitude reduction of time domain channel wave signal after filtering and inverse transformation only with positive frequency time-frequency spectrum. The improved discrete S-transform algorithm proposed in this paper can effectively overcome the time-frequency spectrum distortion of traditional discrete S-transform at negative frequency. By comparing and analyzing the time-frequency spectrum of channel wave seismic data obtained by two discrete algorithms of S-transform, it is verified that the improved discrete S-transform algorithm realizes the symmetry of Gaussian window function, and makes the transformed time-frequency spectrum in negative frequency part no longer distorted and symmetrical with the time-frequency spectrum of positive frequency part. Through applications of time-frequency domain filtering and channel wave signal extraction of numerical simulation and measured channel wave seismic data, it shows that compared with the traditional discrete S-transform, the actual distribution of spectral energy in time domain and frequency domain of the improved discrete S-transform is more consistent with that of the original signal in time domain and frequency domain, amplitude preservation extraction of channel wave signal can be realized by filtering and inverse S-transform in time domain and frequency domain.