Abstract: With the rapid development of technologies such as electromagnetic spectrum operations, the electromagnetic environment has become increasingly complex, and signal aliasing problems have become increasingly prominent. Conventional aliased signal separation typically requires prior knowledge of the number of signals, making it difficult to adapt to unknown open electromagnetic environments. To address the above problems, this paper proposes an integrated detection-and-separation single-channel deep learning model, namely the Electromagnetic Detection-Separation Network (EDSNet). The method first designs a high-dimensional feature processing architecture to achieve representation of aliased signals; secondly, it integrates multi-scale convolution and lightweight attention mechanisms to design a multi-kernel attention encoder, which enhances the feature representation capability for aliased signals; furthermore, a detection head and mask generation module are designed, where the detection head estimates the number of signal components in the aliased signal mixture, and the mask generator adaptively selects the corresponding mask generation units according to the number of signals to achieve reconstruction and separation of each component signal. Simulation experiments show that the proposed method can still effectively achieve signal separation when the number of aliased signals is unknown.