【摘要】 基本ART模型缺乏对样本集拓扑结构及分布特性的学习,导致其抗噪性能较差,容易产生类别增殖现象,进而导致分类性能不稳定。本文将基本ART模型与SOM、GNG的侧向连接和动态拓扑结构相结合,提出了一种具有拓扑保持结构的ART模型(topology preserving ART model,TPART)。利用构建的模型对聚类状分布的高斯分布数据集进行测试,在受到大量孤立噪声点干扰和输入样本顺序的影响下,其性能相对于Fuzzy ART有较大提高。进一步将其应用于灰度图像分割,也取得了较FuzzyART更好的分割结果。更多还原

【Abstract】 As a result of lacking learning of topological structure and distribution characteristic of sample set, the basic ART model is fragile to noise and prone to cause category proliferation, even induce poor classification stability. Combining the basic ART model with SOM and GNG models that possess lateral connection and dynamic topological structure, a topology preserving ART model is proposed. The proposed model is tested on a cluster-shaped Gaussian data set. Compared with Fuzzy ART, the model achieves better performance in the presence of a large number of outliers and different order of input vectors. Furthermore, the model is applied to gray image segmentation and better segmentation results are also obtained.更多还原