Existing techniques primarily rely on traditional time-frequency domain features and neglect the spatial domain features of brain activity. To classify emergency braking intentions and normal driving, a new model integrating multi-scale convolution, brain functional networks, and graph convolutional neural networks was proposed. First, multi-scale convolution was used to extract multi-scale features fused from the time-frequency domain. Then, a brain functional network was constructed based on the brain functional connectivity measurement matrix to obtain spatial graph structure information. Finally, graph convolutional neural networks were employed to integrate multi-scale features and spatial graph structure information for the classification of emergency braking EEG signals. Experimental results showed that the proposed model achieved an accuracy of over 93.00% across multiple subjects in a public dataset, with a maximum of 95.60%. In the single-subject condition, the accuracy exceeded 92.00%, reaching as high as 98.94%. Ablation experiments confirmed that each module significantly contributed to the improvement of the model’s performance. Compared to six existing EEG signal classification algorithms on the same dataset, the proposed model showed superior performance.