Aiming at the problems that there are the insufficient estimation performance and the inadequate channel feature extraction when the existing channel estimation algorithms based on the deep learning (DL) for OFDM systems generate the mismatch of the signal-to-noise ratio (SNR),a channel estimation scheme based on the noise estimation and multiscale channel reconstruction neural network (NE-MCRNNet) is proposed.Firstly,the noise estimation network (NENet) is designed to estimate the SNR of the current transmission environment,and then the multiscale channel reconstruction network (MCRNet) is constructed and the network is selected according to the estimated SNR value,the channel matrix at the pilot frequency is reconstructed by using the multiscale feature extraction for channel reconstruction,the ability of extracting the channel information at different scales is enhanced,finally,the channel estimation is accomplished by using the residual structure to focus on learning the high-frequency differences.The simulation result shows that the proposed channel estimation scheme can obtain the better estimation results than the traditional channel estimation,and can better adapt to the case of the SNR mismatch,so the proposed channel estimation scheme can better meet the channel estimation requirements of OFDM systems.