Shortwave Signal Specification Recognition Based on Feature Fusion Network

‍ ‍In order to solve the problems of single feature selection and weak performance in distinguishing signal with same modulation type in shortwave signal specification recognition， a signal specification recognition algorithm based on feature fusion network is proposed， and a recognition model with vector diagram and signal data stream as network input is designed. Firstly， by signal preprocessing， the vector diagram and standardized signal data matrix are obtained as the feature source， and the signal specification recognition model based on feature fusion network is designed. Secondly， the dense convolution algorithm of the model is used to extract， fuse and learn deep features of the vector diagram and data matrix， while avoiding network degradation， so that the target signal specification recognition is completed. In addition， random frequency offset strategy is designed to further improve the generalization ability in the construction of the shortwave signal data set. Simulation results show that the proposed algorithm has a good recognition effect on signal set with the same modulation type， the model space is small and the operation speed is fast， the recognition accuracy can reach 98% when the signal to noise ratio is 0 dB.