基于筋骨假人和意图标注的躯干肌电预测结果校正

CALIBRATION OF PREDICTION RESULTS OF TRUNK EMG BASED ON MUSCULOSKELETAL MANNEQUIN AND INTENTION LABELING

在人机交互领域,预测躯干肌电信号极具应用潜力。但手部操作、平衡条件等因素会导致躯干肌肉控制模式转换,破坏躯干基于运动信号和肌电信号间的映射关系,因此很难实现高精度躯干肌电预测。为实现对应意图的躯干肌电预测,在设定弯伸腰任务内,测量一组部分椎旁肌肌电信号及运动信号,通过对多维椎旁肌肌电信号的多次两步聚类编码,将其转化为聚类编码号组成的动作向量,作为BiLSTM-CRF算法的输入,实现躯干肌肉动作的分时段标注,进而利用筋骨假人分别校正躯干肌电预测结果。预测校正结果可反映个体特征、躯干和手部动作意图。

Prediction of trunk muscle electromyography(EMG)has great application potential in the field of man-machine interaction.However,the control modes of the trunk muscles alternate with human intentions,hand operations,balance conditions and other factors,which undermines the mapping relationship between motion signals and EMG signals.Therefore,it is difficult to realize the high-precision prediction of the trunk EMG.In order to achieve EMG prediction corresponding to the intentions,the EMG signals of a group of paravertebral muscles and motion signals were measured during preset flexion-extension tasks.The multi-dimensional EMG signals of paravertebral muscles were transformed into action vectors composed of Two-step Clustering numbers.The action vectors were used as the input of BiLSTM-CRF algorithm to realize the tagging of trunk muscle actions during different periods,and the musculoskeletal mannequin was used to calibrate the trunk EMG prediction results.The calibration results can reflect the intention of the trunk,the hands and the individual characteristics.