Study on a Parameter-Adaptive Radar Signal Preselection Algorithm Based on Pulse Feature Importance Assessment

This paper addresses the challenge that radar signal preselection methods in complex electromagnetic environments often depend on manually selected clustering features and require preset initial parameters for clustering algorithms. To overcome these limitations, we propose an improved preselection algorithm that integrates feature optimization with parameter adaptation. First, a feature importance assessment criterion based on information entropy is developed to quantitatively analyze multidimensional pulse descriptor parameters. This criterion selects feature combinations with the highest distinguishability for clustering preselection, thus reducing data redundancy. To mitigate the sensitivity of adaptive parameter extraction to interference data, we introduce the theory of data fields to construct a pulse data potential energy distribution model. By setting dynamic potential thresholds, interference pulses with discrete distributions are effectively removed. In the core preselection phase, we innovatively apply the Kneedle algorithm to detect the knee point in the "k-nearest neighbor distance curve," enabling the intelligent optimization of the DBSCAN algorithm’s domain radius parameter. Simultaneously, based on this optimized value, the spatial distribution characteristics are reanalyzed, and the cumulative distribution function (CDF) is used to derive the optimal threshold for the minimum number of data points. This approach overcomes the limitations of traditional radar signal preselection methods, which rely on empirically set parameters, and eliminates the need for manual intervention. Simulation results demonstrate that the proposed algorithm significantly enhances accuracy compared to parameter-adaptive DBSCAN algorithms and those that depend on empirically set parameters. It achieves fully adaptive parameter generation for radar signal preselection.