基于多特征融合的PSO-SVM优化算法睡眠脑电方法

打开文本图片集

中图分类号：TS976.9 文献标志码：A 文章编号：2097-2911-（2025）06-0001-18

Abstract： In response to the issues of single feature dimension and low efficiency in hyperparameter optimization in existing sleep staging methods，amulti-feature analysis approach that integrates time-frequency domain features with nonlinear dynamic parameters is proposed.The Support Vector Machine （SVM） clasification model is improved by combining the Particle Swarm Optimization （PSO） algorithm to achieve the staging of sleep EEG signals.Through wavelet threshold denoising and Principal Component Analysis （PCA） for dimensionality reduction，the feature dimension is reduced from 15 to 6 （with a cumulative contribution rate of （204号 92.3% ）. A feature set including time-domain statistics， frequency band energy ratios，and nonlinear parameters is constructed.The kernel parameters and penalty factors of the SVMare optimized using PSO to enhance classification performance.In the 5-fold cross-validation on the Sleep-EDF public dataset （10 subjects，4526 segments），an overall accuracy of over 90% isachieved，which is an 8.2 percentage point improvement compared tothe traditional grid search method.The experimentalresults demonstrate thattheproposedmethod，through feature fusionand parameteroptimization，providesareliable technicalsolution for thedevelopmentof clinical sleep monitoring devices.

Keywords：sleep EEG signals；Support Vector Machines;multi-feature fusion; principal component analysis; Particle swarm optimization; sleep classification

睡眠作为人类生命活动的重要周期现象，不仅是生理机能恢复的关键阶段，更是维持认知功能、代谢平衡及免疫调节的复杂神经过程[1-3]。（剩余23540字）