三种谱聚类算法及其应用研究

发布时间：2018-10-24 14:00

【摘要】：为全面理解谱聚类(spectral clustering,SC)算法性能,利用四类几何结构数据,对规范化割(normalized cut,Ncut)、稀疏子空间聚类(sparse subspace clustering,SSC)和谱曲率聚类(spectral curvature clustering,SCC)三种谱聚类算法进行对比分析。结果表明,三种算法的聚类结果各有差异,但每类数据都可以找到相对最有效的聚类算法。Ncut无法处理相交的数据,适用性较差;SSC算法适用性较强,但聚类精度不高;SCC算法具有适用性强、精度高等特点,能够实现四类几何结构数据的有效聚类。此外,改进的SCC算法有效地实现了有数据间断的两条相交螺旋线聚类。最后,分析了现有SCC算法存在的不足,并指出进一步研究的方向。
[Abstract]:In order to fully understand the performance of spectral clustering (spectral clustering,SC) algorithm, three spectral clustering algorithms, normalized cut (normalized cut,Ncut), sparse subspace clustering (sparse subspace clustering,SSC) and spectral rate clustering (spectral curvature clustering,SCC), are compared and analyzed using four kinds of geometric structure data. The results show that the clustering results of the three algorithms are different, but each kind of data can find the most effective clustering algorithm. Ncut can not deal with intersecting data, and the applicability of SSC algorithm is strong. However, the clustering accuracy is not high, and the SCC algorithm has the characteristics of strong applicability and high precision, so it can effectively cluster four kinds of geometric structure data. In addition, the improved SCC algorithm effectively implements the clustering of two intersecting helical lines with data discontinuity. Finally, the shortcomings of the existing SCC algorithm are analyzed, and the direction of further research is pointed out.
【作者单位】：安徽大学资源与环境工程学院;安徽省地理信息工程中心;安徽省农业生态大数据工程实验室;
【基金】：国家自然科学基金资助项目(41501376) 安徽省自然科学基金面上资助项目(1608085MD83,1308085MD52) 地理国情监测国家测绘地理信息局重点实验室开放基金资助项目(2014NGCM05) 数字制图与国土信息应用工程国家测绘地理信息局重点实验室开放基金资助项目(GCWD201406)
【分类号】：TP311.13

【相似文献】