中文摘要：

基于表面肌电信号的肢体动作模式识别是假手仿生控制的基础。为提高动作模式识别率，从肌电信号的产生机理出发，选取分别表征肌电信号形态特征——细节复杂度和整体自似性的近似熵指标和分维数指标，作为模式识别的特征向量；同时提出一种具有增量学习能力的K最近邻（KNN）模型增量学习算法作为模式识别的分类器。在对10位受试者手腕的4个精细动作（腕伸、腕屈、腕内旋、腕外旋）的识别实验中，取得了92．5％以上的正确识别率。同时对增量学习能力对分类器动作模式识别率的影响做对比实验，当假肢使用者生理变化时，以KNN模型增量学习算法作为分类器比采用不具增量学习能力的KNN模型算法的识别率高4．5％。实验表明，该肌电信号动作模式的识别方法方案合理，具有应用价值。

英文摘要：

Action pattern recognition of limbs using sEMG is the basis for bionic control of a prosthetic hand. In consideration of the generation mechanism of sEMG, the approximate entropy and the fractal dimension, which feature the sEMG＇s morphological characteristics including complexity and overall self-similarity, was chosen as the feature vector of pattern recognition to improve action mode recognition rate. In the meantime, a K nearest neighbor （KNN） model incremental learning method with incremental learning ability, was presented as a classifier of pattern recognition. In pattern recognition experiment of classifying four fine movements of the wrist （namely wrist extension, wrist flexion, wrist pronation, wrist supination） with 10 participants, the correct mode recognition rate is above 92.5%. In a contrast experiment that was designed to evaluate the effects of the increment learning ability to the action mode recognition rate, the correct recognition rate is 4.5 percent higher than KNN mode arithmetic without incremental learning ability when the prosthetic users changed physiologically. The above experimental result shows action mode recognition method based on the EMG is reasonable and practical.