基于自组织增量学习神经网络的信息融合技术

Information Fusion Based on Self-organizing Incremental Neural Network

针对传统神经网络在实际信息融合过程中存在的一些缺陷,提出一种基于自组织增量学习神经网络(self-organizing incremental neural network,SOINN)的信息融合方法。对不同类型传感器接收到的异构数据,使用增量式正交分量分析(incremental orthogonal component analysis,IOCA)方法进行数据自适应降维和特征提取,将提取出的不同类型特征输入到SOINN中,根据不同数据类型生成相应的神经元连接区域,建立神经区域间的联想记忆,从而实现在数据层、特征层以及决策层3个层面上的信息融合。实验结果表明:该方法能够实现对机器人传感器采集到的多源异构数据进行自适应降维和自组织学习,形成机器人的决策判断和行为指令。

To solve the problems in information fusion with traditional neural networks, an information fusion method based on self-organizing incremental neural network（SOINN） is proposed. The proposed fusion system can receive the input data with any dimension and any format from different kinds of sensors. The incremental orthogonal component analysis（IOCA） method is used to reduce the dimensionality of data and extract features adaptively. Then the heterogeneous features are learnt by SOINN. During this period, the connected regions of neurons are generated based on the heterogeneous features and the associated relations between the neuron regions are built. By this way, the data fusion is realized at all of the data level, the feature level, and the decision level. It＇s shown from the experiments that the dimension can be reduced and the data recorded by different sensors can be learnt adaptively, and then the decisions and instructions of the robots are generated accordingly.