Abstract: In the orthogonal frequency division multiplexing system of the 5G wireless network, channel estimation plays an important role in realizing accurate signal recovery and reliable communication. The traditional channel estimation method has defects of slow speed and low accuracy. Although the technique based on deep learning can solve this problem, the existing method based on deep neural networks still has the following shortcomings: the traditional convolutional neural network method has low accuracy due to its limited representation capability; although the channel estimation method based on Transformer can achieve high reconstruction quality, it has some problems such as high computational complexity, long time consumption and large number of parameters. To seek a balance among computational complexity, the number of parameters, and estimation performance, a channel estimation method based on wavelet decomposition and dual-stream network is proposed in this paper, aiming to estimate high-resolution channel responses from low-resolution channel responses. To enhance the representation ability of wavelet domain network, a spatial feature extraction module is introduced to assist the estimation of wavelet coefficients with the help of spatial information. In order to achieve the fusion of spatial information and wavelet domain network features, a two-branch network is designed. The obtained wavelet coefficients and spatial features are input into the dual-stream network in four groups for fusion. Simulation results show that the proposed network has low computational complexity and parameters, and can achieve efficient and accurate channel estimation.