The density peak clustering algorithm is simple and efficient, capable of identifying clusters of arbitrary dimensions and shapes, and has been widely applied in various fields. However, this algorithm also has some issues, such as sensitivity to the truncation distance parameter, difficulty in finding the cluster centers of low-density regions, and a tendency to produce the ‘domino effect’. To address these issues, this paper proposes a density peak clustering algorithm with mixed nearest neighbors and multi-cluster merging. First, by comprehensively considering the global distribution and local structure of the samples, natural nearest neighbors and $ k $-nearest neighbors are introduced to redefine local density, eliminating sensitivity to the truncation distance parameter and augmenting the local density of samples in low-density regions to improve the identification of cluster centers. Then, the samples are divided into multiple micro-clusters, and inter-cluster association is utilized for merging to reduce the misallocation of samples that are far from the cluster centers, thereby alleviating the ripple effect of allocation errors. Finally, tests conducted on both synthetic and real datasets demonstrate that the proposed algorithm outperforms its comparative counterparts in overall performance.