现代化信息战争中,多传感器对目标协同感知的需求愈发迫切,如何有效融合传感器的多源异构信息是当前需要研究的问题.针对可见光图像和电磁信号2种异构信源,提出1种基于多分支深度学习网络的信息融合模型:分别基于CNN(Convolutional Neuron Networks)和CLDNN(Convolutional,Long Short-Term Memory,Fully Con-nected Deep Neural Networks)设计独立的的特征学习网络,统一异构信息的特征维度并进行矢量融合,再通过全连接网络对融合后的高维特征离散映射,实现目标分类.仿真结果表明:相较于采用单一传感器信息进行特征描述,该论文设计的信息融合模型能够有效利用目标的多模态信息,提升目标识别性能.
An Object Recognition Method Based on Multi-Source Heterogeneous Information Fusion
In modern information warfare,the demand for cooperative perception of targets through multiple sensors is be-coming urgent.How to effectively fuse multi-source heterogeneous information from sensors is a current issue that needs to be investigated.Aiming at two heterogeneous information sources,including visible images and electromagnetic sig-nals,a multi-source heterogeneous information fusion method based on multi-branch deep learning networks is proposed.Two independent feature learning networks with reference to CNN(Convolutional Neuron Networks)and CLDNN(Con-volutional,Long Short-Term Memory,Fully Connected Deep Neural Networks)are designed.The dimensions of feature vectors extracted from the heterogeneous information are unified and then the vectors are fused.Finally,the fused high-dimensional features are discretely mapped through a fully connected network to achieve target classification.The simula-tion results show that compared with using information from a single sensor for feature extraction,the information fusion model designed in this paper can effectively utilize the multimodal information of targets and improve the recognition per-formance.
国家科技期刊平台
NSTL
万方数据
