Research On Key Technologies Of Blood Pressure Signal Feature Representation

Non-invasive blood pressure signal analysis has broad application prospects in the field of intelligent monitoring of chronic diseases and emotions.However,currently,blood pressure analysis faces problems such as limited user data,large differences in the acquisition of signals by different devices,and high difficulty in building monitoring models,and it is urgent to propose more effective intelligent analysis methods.In view of the above problems,this paper conducts in-depth research from three perspectives:the construction of blood pressure signal feature representation model,the robust characterization of limited blood pressure data,and the analysis of cross-distributed blood pressure characteristics.The main work of the thesis is as follows:1.In the task of solving the difficulty of constructing blood pressure analysis model,the attention mechanism of long sequence is introduced,a parallel self-attention network with interwoven multiple signals is established,and the cross-correlation between signals is extracted to achieve efficient representation of non-invasive blood pressure signals.2.In the task of solving insufficient blood pressure data of users,based on the less-sample learning strategy,a blood pressure data enhancement method based on pseudo-feature generation is proposed,and the diversity of pseudo-features is further expanded by combining Gaussian distribution,which improves the robustness of blood pressure signal representation under few-sample conditions.3.In the task of solving the difference in blood pressure signal distribution across collectors,a new domain transfer strategy is proposed,which simultaneously models the features of source domain data and target domain data through Fourier transform and cluster enhancement algorithm,learns the feature distribution of different domains,and designs different loss functions for regression and classification tasks to constrain model fitting,which improves the generalization performance of blood pressure signal analysis.Through ablation experiments and comparative experiments,the effectiveness of the proposed algorithm is verified on public datasets and private datasets,showing its potential in the field of intelligent analysis of non-invasive blood pressure signals.