Aiming at the low recognition accuracy of radar signals under the low Signal-to-Noise Ratio (SNR), we propose a method based on multi-time-frequency images fusion and dilated residual network.Firstly, radar signals are transformed into different time-frequency images by various time-frequency analysis methods, and these time-frequency maps are fused and processed.Then, a new network model is constructed, which combines the dialated residual network with Distinguishing Feature Fusion Extraction (DFFE) module to identify 10 types of radar signals.Simulation results show that when SNR is -6 dB, the overall recognition accuracy of the proposed method for 10 types of signals reaches 98.7%.