In modern electronic countermeasures systems, accurate detection of unknown user signals and identification of signal modulation mode are essential for signal demodulation and communication countermeasures. Energy detector is a traditional unknown signal detection method, but it has SNR-wall problem, and it cannot distinguish different types of signals. In this paper, to address the problems encountered by traditional methods, we first propose a signal detector based on neural networks. It uses the time domain waveform information of the signal to achieve good detection performance, and it can work without relevant prior knowledge. Furthermore, the effects of different modulation modes and SNR on performance are studied. Finally, a signal classifier based on deep learning is proposed, which can achieve good classification performance for detected signals.