期刊文献+

基于改进变分自编码器的多物理域多模态信息融合方法研究 预览

Multi-physics Multi-modal Information Fusion Method Based on VAE
在线阅读 下载PDF
收藏 分享 导出
摘要 多物理域多模态信息融合具有复杂、高维度、计算量大、鲁棒性差、异构数据处理难的特点,传统基于变分自编码器方法,由于采用单一的信息通道,缺乏对不同模态不同时频域空间信息交换,忽略了不同尺度的特征融合,造成生成结果质量较差。本文在变分自编码器的基础之上,加强了变分自编码器的信息流向,提出基于改进变分自编码器的多物理域多模态信息融合方法,使得融合结果更能反映不同物理域的观测真实性。同时对比了病态数据融合、彩色图像融合深度图像两种不同的多模态信息融合与传统算法效果,在频谱图的定性指标和PSNR/SSIM定量指标上,有明显提升。 Multi-physics multi-modal information fusion has the characteristics of complexity,high dimension,large computation,poor robustness,and difficult heterogeneous data processing. The traditional variation-based self-encoder method uses a single information channel and lacks different modes. The spatial information exchange in different time and frequency states ignores the feature fusion of different scales,resulting in poor quality of the generated results. Based on the variational self-encoder,the information flow direction of the variational self-encoder is strengthened. A multi-physics multi-modal information fusion method based on improved variational self-encoder is proposed,which makes the fusion result reflect different physics. The observational authenticity of the domain. At the same time,the two different multi-modal information fusions and the traditional algorithm effects of ill-conditioned data fusion and color image fusion depth image are compared,and the qualitative indicators and PSNR/SSIM quantitative indicators of the spectrogram are significantly improved.
作者 符晓明 Fu Xiaoming(不详)
出处 《计量与测试技术》 2019年第9期53-55,共3页 Metrology and Measurement Technique
关键词 变分自编码器 信息融合 VAE information fusion
作者简介 符晓明,男,高级工程师。
  • 相关文献
投稿分析

相关作者

内容加载中请稍等...

相关机构

内容加载中请稍等...

相关主题

内容加载中请稍等...

浏览历史

内容加载中请稍等...
;
使用帮助 返回顶部 意见反馈