基于LightGBM的改进表面肌电信号手势识别研究  

Research on Improved Gesture Recognition of Surface EMG Based on LightGBM

作　　者：梅雪峰 赵礼峰[1] MEI Xuefeng;ZHAO Lifeng(School of Science,Nanjing University of Posts and Telecommunications,Nanjing 210023)

机构地区：[1]南京邮电大学理学院,南京210023

出　　处：《计算机与数字工程》2022年第1期95-99,共5页Computer & Digital Engineering

基　　金：国家自然科学基金青年基金项目(编号:61304169)资助。

摘　　要：为了提升表面肌电信号(sEMG)手势动作识别的准确性和训练效率,提出一种基于LightGBM的手势识别模型。传统的GBDT算法训练效率较低,准确率无法快速提升,LightGBM算法采用基于梯度的单侧采样和互斥特征捆绑改进性能,具有训练速度快、占用内存低、分类准确率高的优势。将臂环采集到的8通道sEMG数据按时间顺序进行扁平化处理,提取有效特征。实验结果表明,经过LightGBM改进的sEMG手势识别模型取得较高准确率,并且显著提升训练速度。In order to improve the accuracy and the training efficiency of hand gesture recognition using surface electromyogra⁃phy(sEMG),a hand gesture recognition model based on LightGBM is proposed.The traditional GBDT algorithm has low efficiency of training,which cannot improve the accuracy rapidly.In LightGBM algorithm,gradient-based one-side sampling and exclusive feature bundling are proposed to improve the performance with the advantages of fast training efficiency,low memory consumption and high classification accuracy.The 8-channel sEMG data collected by the armband are flattened in chronological so that extract⁃ing effective features.Finally,the experimental results show that the improved gesture recognition model of sEMG based on LightG⁃BM achieves high accuracy and significantly improves the speed of training.

关 键 词：手势识别 表面肌电信号 LightGBM 基于梯度的单侧采样 互斥特征捆绑 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]