Bayesian Optimization-Based Adaptive Radar Clustering Algorithm

Radar signal sorting is a critical component for achieving battlefield situational awareness in modern electronic warfare. However, in complex electromagnetic environments, the severe overlap of emitter signal features, coupled with substantial random interference, poses significant challenges for precise sorting under non-cooperative conditions. Clustering algorithms are widely applied to separate unknown radar emitter signals due to their unsupervised learning capabilities. To address the strong parameter dependency and unstable performance of existing clustering algorithms—particularly when pulse overlap and heavy interference cause local density distortion—this paper proposes an adaptive Density Peak Clustering (DPC) algorithm based on Bayesian Optimization. The proposed method incorporates a Bayesian global optimization mechanism, utilizing Gaussian Process Regression to establish a probabilistic model mapping hyperparameters (such as cutoff distance and density weights) to multidimensional clustering quality metrics. This allows for the adaptive search of optimal parameter combinations without requiring prior knowledge. Simulation experiments validate the algorithm's performance under pulse overlap and interference conditions. Results demonstrate that the algorithm achieves an average sorting accuracy of 96.6% in the presence of overlapping pulses and 20% interference signals.