The density peaks clustering algorithm(DPC) can identify clusters with arbitrary shapes, but for clusters with multiple density peaks, the DPC may identify multiple cluster centers, leading to wrong cluster partitioning. Therefore, a density peaks clustering algorithm based on low density score(LS-DPC) is proposed. The algorithm firstly uses low density score to enlarge the density difference of data points and reduce the density difference of adjacent regions with large overall density difference, so that the density distributions of all regions in a single cluster are reconstructed into a single-peak density distribution, and then automatically obtains the center centers of sub-clusters according to the low density score. After the sub-clusters are obtained, the sub-clusters are merged according to the density intersection condition to complete the clustering. The proposed LS-DPC algorithm is compared with $k $-Means, SC, DPC, DN, Extreme and ICKDP. Experimental results show that the proposed algorithm outperforms the comparison algorithms on complex datasets and UCI datasets.