In order to solve the problem of incomplete extraction of spatial, temporal and spectral features of EEG information in emotion recognition research, a novel emotion recognition method based on spatial-temporal-frequency multi-domain EEG features using SimAM attention mechanism is proposed. The original EEG data are processed to obtain four-dimensional data feature representations with time, frequency and spatial feature information, and three-dimensional convolutional neural network (3D CNN) and two-dimensional depth separable convolutional network (2D DSCN) are used to learn spatial and frequency domain information features. SimAM attention mechanism is used to further optimize the distribution of weights of neurons in the brain, and BiLSTM network is used to learn the characteristics of temporal information. The experimental results show that the average classification accuracy is 93.43% and 94.21% in the two continuous dimensions of Valence and Arousal, respectively, which provides a new method for EEG emotion recognition tasks and has certain theoretical reference value.