Radar Signal Recognition Method based on Transfer Learning

Radar Signal Recognition Method based on Transfer Learning

연구 분야: Infrastructure

학회: ISIA '23: Proceedings of the 2023 International Conference on Intelligent Sensing and Industrial Automation

Radar signal recognition has a guiding role for electronic warfare systems. The traditional radar signal recognition method confirms the target model through signal feature comparison, which is suitable for electromagnetic environments with single radar signal and low signal feature complexity. With the continuous development of radar technology, traditional radar signal recognition methods have been difficult to adapt to target recognition in complex electromagnetic environments. In order to improve the accuracy of radar signal recognition in complex electromagnetic environments, this paper proposed a radar signal recognition method based on transfer learning. At first, the proposed method performed pre-processing operations on the initial signal data collected by the sensor, including normalization, noise truncation, feature extraction, etc., and established a database for the processed signal data. Afterwards, a feedforward BP neural network model was trained to classify the signal data. Finally, on the basis of the transfer learning, pulse characteristic parameters (radio frequency, etc.) were added to the output layer of the feedforward BP neural network, it served as a new input layer to train deeper network models, making it adaptive and deviation-resistant. The feasibility of the proposed method was verified with six sets of different sample data. The results showed that the average recognition accuracy of the method in this paper reached 93.3%, compared with the traditional ANN-based method, the average recognition accuracy is improved by 12%, which verifies the reliability of the proposed method.