Using electroencephalogram (EEG) signals to infer the content of thoughts in the brain may lead to science fiction-level technologies such as telepathy and mind control, and recognizing imagined digits is the most basic and important task in decoding electroencephalogram signals. However, existing methods for recognizing imagined digits have limitations, as they can only handle with specific tasks or specific datasets. To fill this gap, this paper presents a masked electroencephalography model (MEM) based imagined digit recognition scheme. This enhanced technology has generality and is not limited to specific tasks and datasets. In particular, the MEM model is with a multi-layer self-attention mechanism, which adaptively captures multi-channel EEG features through EEG context inference, transforming imagined digit EEG signals into a unified intermediate vector representation. With the help of this feature representation capability, subsequent multi-layer convolutional neural networks (CNN) can effectively classify these imagined EEG signals, thus enhancing the recognition of imagined digits. The effectiveness of the proposed scheme is validated through verification experiments on three typical imagined digit datasets, including EPOC, MUSE, and INSIGHT. The experimental results indicate that our method can achieve good recognition results on these three datasets, reaching 93.7%, 95.7%, 91.3% respectively.