With the widespread use of various communication, navigation, and remote control devices, the electromagnetic environment has become increasingly complex. Monitoring and controlling the electromagnetic spectrum in a cluttered electromagnetic environment, and accurately identifying unauthorized and non-cooperative devices, has become an urgent need. In recent years, identification technologies based on deep learning have developed rapidly, but there is still a lack in the mechanism of multi-domain feature fusion, and the identification performance under non-cooperative conditions is still relatively weak. In response to the problem of identifying specific electromagnetic target signals in situations with mixed signals, a method for electromagnetic signal identification based on multi-domain feature fusion and target detection architecture is proposed. This method utilizes short-time Fourier transform to achieve the fusion of signal features in the time-frequency domain, and based on the YOLO target detection architecture, a joint feature extraction and recognition mechanism in the time-frequency domain is designed. The algorithm is validated for performance on real-world data and public datasets, and the experimental results show that the recognition accuracy of this algorithm is significantly higher than that of comparative deep learning identification methods.