Abstract: To address the issues of low accuracy in motor imagery electroencephalogram (MI-EEG) decoding models and the difficulty of existing methods in effectively integrating local spatial features with global temporal dependencies, this paper proposes a novel decoding model, EEG-EANet. Firstly, the model extracts temporal features at multiple scales through a dual-branch temporal convolution, introduces an efficient multi-scale attention (EMA) mechanism to enhance the ability of spatial feature extraction, then extracts spatial features via spatial convolution, and subsequently adopts a parallel structure of average pooling and variance pooling to extract multi-modal temporal information. Secondly, it uses a self-attention mechanism to effectively capture global dependencies, and fuses the dual-pooling features through a convolutional encoder to generate discriminative feature representations. Finally, the fused features are input into a classifier to complete the decoding. In the subject-dependent evaluation method, EEG-EANet achieves a classification accuracy of 84.61% and a Kappa coefficient of 0.7948 on the BCI Competition Ⅳ-2a dataset. On the BCI Competition Ⅳ-2b dataset, it reaches a classification accuracy of 88.70% and a Kappa coefficient of 0.774 0. Experimental results demonstrate that the proposed model yields significant performance improvement in MI-EEG decoding tasks and consistently outperforms existing comparative methods.