CAJ | 학술논문

提出一种尺度参数与初始中心自适应的谱聚类算法。与传统谱聚类算法中使用全局尺度参数相比，本算法根据样本数据点邻域信息自适应确定的局部尺度参数，充分考虑了数据的内在结构，并且将粒子群优化的FCM算法引入算法中，寻找最优粒子作为初始中心，解决初始聚类中心敏感性问题。实验结果表明，文章提出的算法相比原始谱聚类算法，聚类结果更稳定，正确率更高。
제출일충척도삼수여초시중심자괄응적보취류산법。여전통보취류산법중사용전국척도삼수상비，본산법근거양본수거점린역신식자괄응학정적국부척도삼수，충분고필료수거적내재결구，병차장입자군우화적FCM산법인입산법중，심조최우입자작위초시중심，해결초시취류중심민감성문제。실험결과표명，문장제출적산법상비원시보취류산법，취류결과경은정，정학솔경고。
This paper presents an improved spectral clustering algorithm where scale parameters and initial centers are adaptiv-ely determined. Comparing with the traditional spectral clustering algorithm using globe scale parameter, the algorithm deter-mines the local scale parameters according to the neighborhood information of sample data adaptively, which fully considers in-herent structure of the sample data. And the particle swarm optimized FCM algorithm is introduced into the algorithm to solve the sensitive problem of initial cluster centers by finding the optimum swarm. Experimental results show that the clustering re-sults of proposed algorithm have more stable and higher accuracy rate compared with the original spectral clustering.