一种用于时空模式识别的综合神经网络

介绍了一种用于时空模式识别的综合神经网络模型，称为ＴＳ－ＬＭ－ＳＯＦＭ．该网络的顶层是一种称为ＴＳ（ｔｅｍｐｏｒａｌｓｅｑｕｅｎｃｅ）的单层时序识别网络，可以把时序模式转换成抽象的空间模式．该网络的底层是ＳＯＦＭ（自组织特征映射网络），用于空间模式特征检测．ＬＭ（学习矩阵）用于上述两层的联接．在实验中，用移动机器人超声阵列传感器作为输入训练，结果表明，该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．

A Hybrid Neural Network for Spatiotemporal Pattern Recognition

A hybrid network is presented for spatio-temporal feature detecting, which is called TS-LM-SOFM. Its top layer is a novel single layer temporal sequence recognizer called TS which can transform sparse temporal sequential pattern into abstract spatial feature representations. The bottom layer of TS-LM-SOFM, a modified self-organizing feature map, is used as a spatial feature detector. A learning matrix connects the two layers. Experiments show that the hybrid network can well capture the spatio-temporal features of input signals.