Signal Recognition Based on Neural Network with Dual-channel Time-frequency Characteristics

With the advancement of communication technology, the utilization of the electromagnetic spectrum has become increasingly sophisticated. However, challenges remain, such as the low recognition rate of signal modulation methods and the tendency for confusion when extracting single time-frequency features. To address issues, this study proposes time-frequency transformation features of the signal's imaginary part and designs a dual-channel neural network Each channel extracts distinct time-frequency features, enhancing the network's capacity to learn complex representations. The extracted channel features are fused in the fully connected layer to improve recognition accuracy. Simulation results indicate that the proposed dual-channel network significantly improves classification accuracy by effectively integrating multi-dimensional time-frequency information, compared with single-channel time-frequency methods and existing image-based approaches.