Application of Clustering and Multi-cluster Selection in SoftMan´s Perception System

SoftMan, the virtual robot in network environment, is a kind of software Artificial Life living in computer networks. It has a humanoid structure, and simulating human, its perception system should be able to recognize perceptive objects. This paper proposes a cooperative classification model for SoftMan´s perception system, in which, different humanoid senses are simulated by multiple classifier systems (MCS). An essential issue in the research of MCS is the method of multiple classifier combination. Accordingly, this paper proposes a method of multiple classifier combination (called CMS), which improves the classification performance by clustering and multi- cluster selection technique. Simulated experiments validate the basic feasibility of the model and the validity of CMS method.