基于多尺度排列组合熵的助行机器人运动相容性识别

Motion compatibility recognition of walk-aid robot based on multi-scale permutation entropy

背景:目前因受到后天环境影响或由于身体功能的退化,下肢肢体运动能力损伤老年人数目持续增长,而老年人下肢肢体运动能力的下降,使得下肢助行机器人的设计与研发成为了当下社会养老护理的重点之一。目的:当助行机器人的运动轨迹与人体期望运动轨迹不一致时,为实现助行机器人可自适应调整,对穿戴助行机器人出现步幅过大不相容、步幅过小不相容及步幅相容3种人-机运动相容性进行模式识别。方法:针对表面肌电信号的非线性、噪声强等特点,提出一种基于小波分解的多尺度排列组合熵方法,对人-机运动相容性的3种情况(步幅过大不相容、步幅过小不相容及步幅相容)进行模式识别。首先,对采集到的表面肌电信号进行小波分解;然后,对肌电信号的各个尺度求取排列组合熵;最后,利用高斯核支持向量机进行模式识别。结果与结论:经小波分解后,在d5尺度上的信号能取得较好的识别率,其中对步幅过大不相容的识别率为92%,步幅过小不相容的识别率为90%,步幅相容的识别率为94%,平均识别率为92%,比原始肌电信号的平均识别率高出了4.67%,且相较于其他常用的多尺度方式也能取得更好的效果。所以在人-机运动相容性识别中,可以将表面肌电信号进行小波分解后提取各尺度上的排列组合熵,有助于提高识别精度。

BACKGROUND:At present,the number of older adults with impaired limb motor ability continues to increase due to the influence of the environment or the deterioration of bodily functions.The decline of the lower limbs'motor ability of the older adults has made the design and development of walk-aid robot become one of current social care focus.OBJECTIVE:When the motion trajectory of walk-aid robot is inconsistent with the desired trajectory of the human,the adaptive adjustment of walk-aid robot should be realized.This study was to recognize three kinds of situations,such as too large stride,too small stride and appropriate stride.METHODS:In view of the nonlinearity and strong noise of surface electromyography,a multi-scale permutation entropy method based on wavelet decomposition was proposed to recognize three situations of man-robot’s motion compatibility.First,wavelet decomposition of collected surface electromyography signal was performed.Then,the permutation entropy was calculated for each scale of surface electromyography.Finally,Gaussian kernel support vector machine was used for pattern recognition.RESULTS AND CONCLUSION:After wavelet decomposition,the signal on the d5 scale could be better recognized.Among them,recognition rate was 92%for too large stride,90%for too small stride,and 94%for appropriate stride.The average recognition rate was 92%,which was 4.67%higher than that of original surface electromyography.Wavelet decomposition also achieved better results than other commonly used multi-scale methods.Therefore,in the man-machine motion compatibility recognition,the surface electromyography signal can be decomposed by wavelet to extract the permutation entropy on each scale,which can help to increase the recognition accuracy.