摘要
为提高基于表面肌电图(surface Electromyo Graphy,sEMG)手势识别的准确率,提出一种改进深度森林相结合的手部运动识别方法.将极致梯度提升(eXtreme Gradient Boosting,XGBoost)树引入深度森林模型,与随机森林和完全随机森林共同组成深度森林的级联结构.深度森林模型在每个层次上集成3种不同的基于树的分类器,共4个决策森林,包括1个随机森林、1个极端随机森林和2个极致梯度提升树,利用不同学习算法之间的互补性来提高分类性能.为评估该模型性能,采集4名健康受试者的表面肌电信号进行手部动作识别验证试验,并与随机森林、支持向量机、一维卷积神经网络及二维卷积神经网络等算法比较.结果表明,提出方法对16种常用手部动作的平均识别精度为94.14%,对表面肌电信号实现了较高的分类准确率.
In order to improve the accuracy of gesture recognition based on surface electromyography(sEMG),an improved deep forest combined hand motion recognition method was proposed.The extreme gradient boosting(XGBoost)tree was introduced into the deep forest model to form the cascade structure of deep forest together with the random forest and the complete random forest.The deep forest model integrates three different tree-based classifiers at each level,a total of four decision forests including a random forest,an extreme random forest and two extreme gradient boosting trees.The classification performance was improved by using the complementarity between different learning algorithms.In order to evaluate the performance of the model,the sEMG signals of 4 healthy subjects were collected for the verification experiment of hand action recognition,and compared with random forest,support vector machine,one-dimensional and two-dimensional convolutional neural networks algorithms.The result shows that the average recognition accuracy of the method for 16 commonly used hand actions is 94.14%,and the classification accuracy of sEMG signals is high.
作者
黄圣
茅健
HUANG Sheng;MAO Jian(School of Mechanical and Automotive Engineering,Shanghai University of Engineering Science,Shanghai 201620,China)
出处
《上海工程技术大学学报》
CAS
2023年第2期190-197,共8页
Journal of Shanghai University of Engineering Science
关键词
深度森林
表面肌电图
深度学习
模式识别
极致梯度提升
动作分类
deep forest
surface electromyography(sEMG)
deep learning
pattern recognition
extreme gradient boosting(XGBoost)
action classification
