论文部分内容阅读
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.
This paper presents an integrated neural network model for spatio-temporal pattern recognition, called TS-LM-SOFM, which is a single layer temporal integration network called Temporal Sequence (TS) As input, due to the excitatory attenuation effect of neurons, the temporal pattern of storage memory gradually expands and transforms into an abstract spatial pattern.The bottom of the model is SOFM (Self-Organizing Feature Map), whose role is to integrate spatial patterns And the actual signal detection.LM (Learning Matrix) is used as the intermediate transition layer between TS and LM.Using TS-LM-SOFM to process the data collected by ultrasonic navigation robots, experiments show that TS-LM-SOFM neural network output The model can represent the spatiotemporal features of the input signal well.