采用广义似然比检测的肌肉收缩起始时刻判断算法

肌电假肢利用残肢残存肌肉的肌电信号实行对假肢的控制。对于低信噪比的残肢表面肌电,本研究采用广义似然比检测方法判断肌肉收缩起始时刻,其中判别阈值与肌电信号信噪比有关。针对不同信噪比的模拟肌电信号,采用离线仿真方法得到肌肉收缩起始时刻检测误差最小的判别阈值,得到信噪比-经验阈值拟合曲线,确定信噪比与阈值的对应关系;根据肌电信噪比由阈值拟合曲线得到判别阈值,采用似然比检测算法在线分析肌肉收缩的起始时刻。与传统算法比较,对于模拟肌电信号,本算法误差均值和标准差分别减小35%和43%;对于真实肌电信号,误差均值和标准差分别减少29%和23%。可见在小信噪比条件下广义似然比检测算法判断肌肉收缩起始时刻较传统算法更为准确。

An Algorithm for Detecting the Onset of Muscle Contraction Based on Generalized Likelihood Ratio Test

The surface electromyography(sEMG) of stump in the amputee is often applied to control the action of myoelectric prosthesis.According to the sEMG signals with low Signal to Noise Ratio(SNR) recorded from the stump muscle,a generalized likelihood ratio(GLR) method was proposed to detect the onset of muscle contraction,where a decision threshold was related with the SNR of sEMG signals,an off-line simulation method was used to determine the relationship between them.For the simulated sEMG signals with a given SNR,the different thresholds were tested,the optimal threshold could be obtained when the detection accuracy was optimized.As a result,the fitted curve was achieved to describe the relationship of the SNR and the decision threshold.Then,the sEMG signals are analyzed on-line by the GLR test for the onset detection of muscle contractions,while the decision threshold corresponding with the SNR was chosen based on the fitted curve.Compared with the classical algorithms,with the simulated sEMG traces,the error mean and standard deviation for estimating the muscle contraction onset were reduced at least 35% and 43% respectively;based on the real EMG signals,the error mean and standard deviation of the onset estimate were separately not less than 29% and 23%.Therefore,the proposed algorithm based on GLR test for the onset detection of muscle contraction was more accurate than other methods,while the SNR of sEMG signals was low.