基于PSO-LSTM模型的上肢动作识别方法

Upper Limb Motion Recognition Method Based on PSO-LSTM Model

针对上肢肌音信号(Mechanomyography,MMG)动作识别准确率不高的问题,提出一种基于粒子群算法(PSO)与长短期记忆网络相结合的混合模型(Particle Swarm Optimization-Long Short Term Memory,PSO-LSTM)的动作识别方法。采用5通道传感器对受试者进行上肢肌音信号采集,使用巴特沃斯滤波(Butterworth Filter)等方法对肌音信号进行预处理,并进行特征提取;构建基于PSO-LSTM的上肢肌音信号识别模型并进行模型训练和测试;最后从不同测度对比了长短期记忆(LSTM)模型、麻雀搜索算法(Sparrow Search Algorithm,SSA)优化的LSTM模型(Sparrow Search Algorithm-Long Short Term Memory, SSA-LSTM)以及PSO-LSTM模型的实验结果。结果表明,PSO-LSTM模型的准确度均高于LSTM、 SSA-LSTM模型,达到96.9%左右,在迭代损失、迭代速度等方面也优于LSTM、SSA-LSTM模型,从而证明了该模型用于上肢肌音信号识别的优越性。

Aiming at the problem of poor recognition accuracy of upper limb muscle sound signal(MMG),A hybrid model(PSO-LSTM)based on Particle Swarm Optimization-Long Short Term Memory(PSO-LSTM)is proposed for action recognition.ADXL357,NI-9202 and other hardware were used to collect 5-channel upper limb muscle sound signals from healthy upper limb subjects,and then Butterworth Filter,Z-scores standardization and other methods were used to pre-process the muscle sound signals.The energy method of sliding window was used for effective action division and feature extraction.The LSTM neural network optimized by PSO algorithm was constructed,and the upper limb muscle sound signal recognition model was trained and tested on the PSO-LSTM model.Finally,we compared the Long Short Term Memory(LSTM)model and Sparrow Search Algorithm from different measures.Experimental results of Sparrow Search Algorithm-Long Short Term Memory(SSA),Sparrow Search Algorithm-Long Short Term Memory(SSA-LSTM)and PSO-LSTM.The results show that the accuracy and recall rate of the PSO-LSTM model are higher than those of the other two models,and the recognition accuracy of the test set is about 96.9%.In addition,the PSO-LSTM model is superior to the other two models in terms of iteration loss and iteration speed,which proves the superiority of this model for upper limb muscle sound signal recognition.