A Novel Pulse Stream Representation and Affinity Measurement Strategy for Radar Recognition With Distribution Pattern Mining

The dynamic and agile pulse stream emitted by modern multifunction radar (MFR) poses a great challenge to its recognition. To solve this problem, we develop a novel pulse stream representation and affinity measurement strategy for radar recognition with distribution pattern mining. To this purpose, we construct a fine-grained self-adaptive histogram algorithm to estimate the distribution probability density of the pulse parameter sequence and further represent it through vector embedding. Based on it, we comprehensively use the distribution features of parameter sequences and their first-order difference sequences to hierarchically represent the pulse stream patterns and construct the radar database for recognition. Finally, the recognition problem is formulated as measuring the distributions’ affinity between the testing pulse stream and each class center in the database. An affinity measurement strategy with the weighted Jensen-Shannon divergence (WJSD) is further proposed to facilitate recognition. Experimental results show that our method achieves superior recognition performance. In addition, our method can be easily applied and extended in practical applications without relying on complex training processes and big datasets.