Abstract: Aiming at the problem of low accuracy of traditional channel estimation algorithms in the harsh environment of underground coal mines, this paper proposes an improved Super Resolution Convdutional Network (SRCNN) for channel estimation. In the improved SRCNN model, the estimated value at the pilot frequency is used as input, and the improved SRCNN model replaces the interpolation process in the traditional channel estimation algorithm to reduce the complexity, and the attention mechanism ECA module is added to improve the learning of channel features to achieve more accurate channel estimation for the underground coal mine environment. Simulation results show that the channel estimation algorithm of the improved SRCNN model outperforms the traditional channel estimation algorithm and improves the estimation accuracy by one order of magnitude compared with the channel estimation of the SRCNN model.