一种时空模式识别的综合神经网络模型

A HYBRID NEURAL NETWORK MODEL FOR SPATIO-TEMPORAL PATTERN RECOGNITION

本文提出一种用于时空模式识别的综合神经网络模型,称为TS-LM-SOFM.该模型高层是一种单层时序整合网络,称为TS(Temporal Sequence)网络.TS网络以稀疏激励模式作为输入,由于神经元的兴奋性衰减作用,存储记忆的时序模式会在空间上逐渐展开,变换为抽象的空间模式.该模型底层是SOFM(Self-Organizing Feature Map),其作用是空间模式整合与实际信号的特征检测.LM(Learning Matrix)作为TS与LM的中间过渡层.利用TS-LM-SOFM对超声导航的机器人实际采集的数据进行处理,实验表明,TS-LM-SOFM神经网络输出的模式能够较好地抽象表示输入信号的时空特征.

In this paper a hybrid network is presented for Spatio-Temporal pattern recognition (STPR) called TS-LM-SOFM. A single layer temporal sequence recognizing network called TS (Temporal Sequence) is introduced as the top layer of TS-LM-SOFM. The temporal sequence of input sparse patterns can spread out within TS layer and achieve abstract feature patterns. The bottom layer is a modified SOFM (Self-Organizing Feature Map) used as a spatial pattern recognizer for feature detecting. LM(Learning Matrix) is used to connect the two layers. In experiment, some mobile robot's sonar sensor data are used for training. Simulations show that the hybrid network can well capture the spatio-temporal features of input signals.