An Improved K-Means Algorithm for Radar Signal Sorting

For radar signal sorting based on pulse descriptors, the inherent limitations of the traditional K-means algorithm include the requirement of a predefined number of clusters, the sensitivity to initial cluster centroids, and easily falling into local optima. To improve the performance of radar signal sorting, an enhanced K-means algorithm is proposed in this paper, which is promising in overcoming limitations mentioned above. The core idea of the proposed algorithm is to obtain the number of cluster centroids by combining the first integer neighbor clustering hierarchy clustering (FINCH) algorithm with the contour coefficient method, select the initial cluster centroids via Kmeans++, identify multiple substitute centroids through the power k-means approach, and enhance clustering performance by reinitializing the centroids in subsequent iterations based on particle swarm optimization rules. The algorithm can automatically obtain the number of clusters and initialize the cluster centroids. Simulation experiments demonstrate that the improved algorithm enhances the applicability of the K-means algorithm and is applicable to the sorting of unknown radar signals.