Spectral clustering (SC) transforms the dataset into a graph structure, and then finds the optimal subgraph by the way of graph-partition to complete the clustering. However, SC algorithm constructs the similarity matrix and feature decomposition for overall datasets, which needs high consumption. Secondly, k-means is taken at the clustering stage and it selects the initial cluster centers randomly, which leads to the unstable performance. Thirdly, SC needs prior knowledge to determine the number of clusters. To deal with these issues, we propose a robust spectral clustering algorithm based on grid-partition and decision-graph (PRSC) to reduce the amount of calculation and improve the clustering efficiency. In addition, a decision-graph method is added to identify the cluster centers quickly to improve the algorithm stability without any prior knowledge. A numerical experiments validate that PRSC algorithm can effectively improve the efficiency of SC. It can quickly obtain the stable performance without any prior knowledge.

光谱聚类（SC）将数据集转换为图结构，然后通过图分区的方式找到最佳子图以完成聚类。但是，SC算法为整个数据集构造相似度矩阵和特征分解，这需要消耗大量资源。其次，在聚类阶段采用k均值，随机选择初始聚类中心，从而导致性能不稳定。第三，SC需要先验知识来确定集群数量。针对这些问题，我们提出了一种基于网格划分和决策图（PRSC）的鲁棒频谱聚类算法，以减少计算量，提高聚类效率。此外，添加了决策图方法以快速识别聚类中心，以提高算法稳定性，而无需任何先验知识。数值实验验证了PRSC算法可以有效提高SC的效率。无需任何先验知识即可快速获得稳定的性能。