Abstract: During operation, information processing terminals such as computers inevitably generate electromagnetic radiation through legitimate covert communication channels. This radiation not only carries terminal-specific useful information, which poses a security risk of information leakage, but may also contains characteristic signals of device operation, presenting a challenge to the accurate monitoring of time-domain electromagnetic anomalies. To address such time-domain electromagnetic leakage signals, a neural network algorithm combining one-dimensional convolution and long short-term memory is proposed. By fusing spatial features and temporal correlation features, the algorithm enhances the deep representation capability of the target signals, enabling intelligent identification of covert-channel electromagnetic leakage and accurate capture of time-domain electromagnetic anomalies from terminals. Experimental results demonstrate that the proposed method achieves an information recognition accuracy of over 84.7%, which can effectively improve the capability of identifying and monitoring time-domain electromagnetic signals from information processing terminals.