采用组合特征法的极限学习机多手势精准识别  

Multi-Gesture Accurate Recognition of Extreme Learning Machine Using Combined Feature Method

作　　者：来全宝 陶庆[1] 胡玉舸 孟庆丰 LAI Quan-bao;TAO Qing;HU Yu-ge;MENG Qing-feng(School of Mechanical Engineering,Xinjiang University,Xinjiang Urumqi 830047,China)

机构地区：[1]新疆大学机械工程学院,新疆乌鲁木齐830047

出　　处：《机械设计与制造》2023年第7期182-186,191,共6页Machinery Design & Manufacture

基　　金：国家自然科学基金资助项目(51865056)。

摘　　要：为了提高手部动作的识别率与响应速度。提出综合特征选择与排列组合的组合特征法并与极限学习计算法(ELM)相结合的多手势模式精准识别方法。首先,运用肌电传感器采集八种手势动作;进而运用去噪技术与起止点检测技术对肌电信号进行预处理;其次,分别提取肌电信号时域、频域、时频域、4阶AR系数和非线性特征,将组合特征法与皮尔森相关系数法和主成分析法(PCA)选优的特征集进行对比;最后,用所选特征集与最优滑动窗相结合,运用极限学习机、神经网络(BP)和支持向量机(SVM)算法进行手势分类。实验结果表明,结合组合特征法与最优滑动窗口设计的ELM算法模型最优,平均识别率高达97.1%,结果超BP算法17.02%,且具有最短的训练与测试时间,有效证明所提方法的精准性和实时性。In order to improve the recognition rate and response speed of hand movements.A combined feature method of comprehensive feature selection and permutation and combination and a precise recognition method of multi-gesture patterns combined with the extreme learning machine algorithm(ELM)is proposed.First,used the EMG sensor to collect eight gestures;then used the denoising technology and the start and end point detection technology to preprocess the EMG signal;Second,extracted the EMG signal in time domain,frequency domain,time-frequency domain,and 4th order AR Coeficients and non-linear features.Compared the combined feature method with Pearson's correlation coefficient method and principal component analysis(PCA)selected feature sets;finally,used the selected feature set and the optimal sliding window to combine and apply extreme learning Machine,neural network(BP)and support vector machine(SVM)algorithms for gesture classification.The experimental results show that the ELM algorithm model combined with the combined feature method and the optimal sliding window design is the best,with an average recognition rate of 97.1%,and the result exceeds the BP algorithm by 17.02%,and has the shortest training and testing time,which effectively proves the proposed method accuracy and real-time.

关 键 词：表面肌电信号 特征值提取 极限学习机 主成分析 多手势识别 

分 类 号：TH16[机械工程—机械制造及自动化] TP391.4[自动化与计算机技术—计算机应用技术]