摘要
通过变分模态分解-希尔伯特变换得到希尔伯特谱能较为准确地反映非平稳信号瞬时频率随时间变化的能量分布。但现有文献多忽略负的瞬时频率,采用正的瞬时频率作为特征,导致希尔伯特谱损失较多有效信息。此外,短时人体行为信号仅包含行为周期的局部信息,这增加了提取时频特征的难度。基于此,文章提出一种新的加权希尔伯特谱的双流卷积神经网络模型。具体地,将短时人体行为信号正/负瞬时频率所对应的希尔伯特谱(H_(pos)/H_(neg))作为WHTs-CNN的输入,H_(pos)通过卷积充分利用希尔伯特谱中的主要信息,更好地捕捉人体行为的动态过程,H_(neg)通过权重模块对H_(neg)进行多次加权处理,对希尔伯特谱的关键信息进行补充和增强,提供更为细致的时频信息。该模型在PAMAP2及自采集数据集上进行实验,结果表明分别具有1.61%、1.31%的性能提升。
The Hilbert spectrum obtained through Variational Mode Decomposition-Hilbert Transform can more accurately reflect the energy distribution of the instantaneous frequency of a non-stationary signal over time.However,most existing literature often ignores the negative instantaneous frequency and uses only the positive instantaneous frequency as a feature,resulting in a significant loss of valuable information in the Hilbert spectrum.In addition,short-term human activity signals contain only partial information about the activity cycle,which increases the difficulty of extracting time-frequency features.Based on this,the paper proposes a new weighted Hilbert spectrum-based dual-stream convolutional neural network model.Specifically,the Hilbert spectrum corresponding to the positive/negative instantaneous frequency of short-term human activity signals(H_(pos)/H_(neg))is used as the input to the WHTs-CNN.H_(pos)fully utilizes the main information in the Hilbert spectrum through convolution,better capturing the dynamic process of human activity.H_(neg)performs multiple weighted processing on H_(neg)through the weighting module,supplementing and enhancing the key information of the Hilbert spectrum,providing more detailed time-frequency information.The model was experimented on the PAMAP2 and self-collected datasets,and the results show performance improvements of 1.61%and 1.31%respectively.
作者
罗彪
宗文杰
陈欣
刘文瑶
齐平
LUO Biao;ZONG Wen-jie;CHEN Xin;LIU Wen-yao;QI Ping(School of Mathematics and Physics,Anqing Normal University,Anqing Anhui 246133,China;School of Computer and Information,Anqing Normal University,Anqing Anhui 246133,China;School of Mathematics and Computer,Tongling University,Tongling Anhui 244061,China)
出处
《铜陵学院学报》
2025年第2期91-96,共6页
Journal of Tongling University
基金
安徽省高校优秀科研创新团队“铜基材料数字化智能制造”(2023AH010056)
铜陵学院联合培养研究生创新基金项目“基于函数型数据分析方法的下肢假肢意图识别”(24tlcb01)
铜陵学院联合培养研究生创新基金项目“基于足底动力相的智能下肢假肢意图识别”(23tlcx01)。
