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

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:: Since the EM algorithm is not suitable for the spatial information requirements of spatial clustering, and although the neighborhood EM algorithm combines the spatial penalty term, NEM requires a lot of iteration in the E-step step. In order to meet the requirements of spatial information and avoid excessive computation, this paper proposes a hybrid incremental NEM algorithm combining EM and NEM, in which the algorithm first performs EM training in random subsamples until the likelihood judgment conditions decrease, the samples are updated according to the incremental factor, and then the samples are turned to NEM training once, so that the cyclic increasing cross-training is carried out, so that the calculation amount is reduced and the performance is improved. The experimental results show that MNEM only needs fewer operations to achieve convergence, and the clustering quality results are better than NEM.