基于表面肌电（sEMG）的人体主动运动意图识别方法是实现带有生物信息反馈控制的人机交互系统的关键理论与技术之一，其准确度和稳定性直接关系到控制效果的优劣。运动神经肌肉系统是非线性动力学系统，故sEMG具有多时间尺度迭加及混沌特性。常用识别方法多对sEMG进行整体分析，无法全面利用其各分量特性，制约了在线识别效果。本研究针对sEMG－肢体运动模式在线识别问题，探讨对sEMG多时间尺度分解及混沌信号短时盲分离的有效方法，提出隔离模式下的特征提取及融合方法；提出建立描述sEMG－肢体运动模式关系的隐马尔可夫模型数据库，基于最大似然估计法探讨模型最优匹配及在线自学习法，建立具有自适应能力的肢体运动意图识别方法；揭示运动过程中组成sEMG的多时间尺度信息和混沌源信息对肢体运动的影响规律，为sEMG－肢体运动在线识别方法提供理论依据，对人机交互、康复机器人、助老助残机器人等领域的研究奠定理论基础。

Surface Electromyogram(sEMG) based human motion intention recognition method is one of the key theories and technologies for human machine interaction systems that achieve biological information feedback control, and its accuracy and stability directly influence the control effect. Motor neuromuscular system is a nonlinear dynamic system, hence sEMG shows multi-timescale and chaos features. The commonly used methods analyze the integration of sEMG, not the features of every component, and thus limit the effect of online recognition. This study aims at sEMG - limb motion intention online recognition method, investigates the effective multi-timescale decomposition and chaos signal blind source fast separation methods, then proposes the feature extraction method in separate mode; and also proposes to build Hidden Markov Model database to describe the relationships between multi-channel sEMG and limb motion patterns, then investigates the optimized model marching and online self-learning method, so as to build sEMG - limb motion intention recognition method of self-adaptive ability. This study can explain the rules of how the multi-timescale features and chaos source signals of sEMG influence limb movements during motion processes, and provides theoretical foundation for sEMG - limb movement online recognition method. Therefore, the results of this study can develop the research works of human-machine interaction, rehabilitation robot and power assist robot for elders and disable.