一种混合递增NEM的空间聚类算法-《郑州大学学报(工学版)》

文章信息/Info

Author(s):: JIA Junjie; ZHANG Qin; Institute of Geology Engineering and Geomatics，Chang’an University，Xi’an 710064，China

Keywords:: spatial clustering; NEM algorithm; Gaussian mixing; Space penalty items

摘要:: 由于EM算法不适合空间聚类对空间信息的要求,而邻域EM算法虽然结合了空间惩罚项,但是NEM在E-step步需要大量的迭代.为了既能满足空间信息的要求,又能避免过多的计算量,本文提出了EM与NEM二者相结合的混合递增NEM算法,算法首先在随机子样本中进行EM训练,直到似然判断条件下降,根据增量因子进行样本更新,然后样本转向NEM训练一次,如此进行循环递增的交叉训练,使得计算量降低,性能提高.实验结果显示,MNEM只需要较少的运算便可达到收敛,聚类质量结果优于NEM.

Abstract:: EM algorithm is inappropriate for spatial clustering which requires consideration of spatial information．Although neighborhood EM algorithm incorporates aspatial penalty term．it needs more iterations in everyE—step．To incorporate spatial information and avoid too much additional computation，this paper proposedmixed increasing NEM algorithm that combines EM and NEM．In MNEM．algorithm first train data based on ran．dora sub—sampling in EM till the likelihood—judgement condition begins to decrease，and update sub—sam．piing．Then training is turned to NEM and runs iteration of algorithm once．Because of this cross train of cycle，MNEM algorithm’computational complexity is decreased and capability is advanced．Experimental result8 8howthat less passes are needed in MNEM to converge and the final clustering quali竹is better than standard NEM．