Multi-channel EEG signals often contain time-frequency overlapping artifacts, and the active components of the EEG signal are easily lost when the artifacts are removed by traditional methods, especially when the frequency characteristics of the artifacts are similar to the EEG signals. In order to solve this problem, an improved artifact removal method based on discrete wavelet transform and independent component analysis was proposed. In this method, the time-frequency decomposition of the EEG signal is first carried out by discrete wavelet transform, the time-frequency characteristics of the signal are captured, and the key signal components are screened out by combining multi-dimensional feature selection and optimization techniques, and finally the FastICA algorithm is used to remove the ocular electrophrasal artifact. In particular, this method optimizes the problem of time-frequency overlap artifacts, and avoids the signal loss caused by traditional methods when the frequency of artifacts and EEG signals are similar by dynamically adjusting the feature selection and filtering strategies. Compared with the traditional method, the signal-to-noise ratio is increased by 4.5 dB, and the power ratio of alpha wave and beta wave is significantly enhanced, which proves that the quality of EEG signal can be effectively preserved while removing artifact, which provides a more efficient and reliable signal processing scheme for EEG devices.