Abstract: For the problem that signal detection is limited by low SNR(Signal-to-Noise Ratio) in complex electromagnetic environment, based on the integration of signal and noise, with the background of all electromagnetic radiation signals in electromagnetic space and deep learning algorithm, a signal detection method is proposed. First, the electromagnetic environment model of the dynamic scene is established, including communication base station signals, radar signals, interference signals, etc. Second, the energy distribution characteristics of the electromagnetic signal in the time-frequency domain are extracted with the Gaussian window Fourier transform. Finally, the convolutional neural network is used for feature selection and classification to achieve the purpose of signal detection. The simulation results show that this method alleviates problem of signal detection limited by SNR to a certain extent, overcomes the defects of traditional energy detection methods and SVM(Support Vector Machines)-based detection methods, and improves the detection performance of electromagnetic signal under low SNR.