A Radar Signal Sorting Method Based On Image Semantic Segmentation

In today's information warfare,electronic countermeasures are increasingly important,and the most critical links is electronic reconnaissance,which is the cornerstone in electronic warfare's attack and defense.In electronic reconnaissance,radar signal sorting is crucial,and it is a prerequisite and guarantee for the entire radar reconnaissance system to work.With the development of radar communication technology,there are more types of radar and advances in radar signal jamming and anti-jamming technology,resulting in an increasingly complex electromagnetic environment where the radar reconnaissance system is located.How to sort out different radar signal sources from the dense radar pulse stream has become a technical problem in the field of electronic reconnaissance.Radar signal contains a lot of parameters,such as inter-pulse and intra pulse parameters.They all play an essential role in radar signal sorting since PRI(pulse repetition interval)is stable in describing most radar pulse signals.Different PRI types have different radar functions.The binned PRI can tell the threat level of the radar radiation source,so most current algorithms are based on PRI for radar signal sorting.The main methods of PRI-based radar signal sorting are divided into two categories.The first category is the traditional,PRI-based radar signal sorting algorithms,using the principle of statistics,these methods are computationally intensive and can easily lead to sorting errors;the second category is based on clustering algorithm sorting,but this method requires setting the number of clusters in advance,i.e.,the number of radar types is known in advance,which is not realistic in the actual radar signal sorting process.For the existing problems of radar signal sorting algorithm and the advantages of using PRI parameters for radar signal sorting,In this paper,two methods of radar signal sorting using PRI parameters are designed based on the image semantic segmentation and the timing information of radar signals.The first method uses the improved U-Seg Net image semantic segmentation network to construct the PRI frequency matrix using TOA parameters.It then uses the proposed improved U-Seg Net network model for radar signal sorting on the PRI frequency matrix.The second method is designed because the TOA sequences of radar signals have a temporal relationship and use an improved Bi-LSTM neural network for radar signal binning.The network has two subnets,the CNN subnet and the Bi-LSTM subnet.It uses the CNN subnet to extract the high-dimensional feature information and the Bi-LSTM subnet to learn the temporal relationship between the radar pulse signals.Then feature fusion is performed to predict the signal type.Through experimental comparison and analysis,it is verified that the binning effect of the two models proposed in this paper is superior in accuracy and faster in prediction compared with the traditional radar signal binning.The improved U-Seg Net proposed in this paper is more robust than the improved Bi-LSTM and performs better in complex radar signals,and the radar signal sorting is faster.