利用柔性神经树的实时肌电信号手势识别模型

Real-time electromyography hand gesture recognition model based on flexible neural trees

传统的手势识别通常采用数字信号处理(Digital Signal Processing,DSP)芯片或者集合方法(Ensem-ble Methods)研究实时识别问题。这些方法易导致数学模型参数繁多、硬件连接复杂和实时识别率较低。提出一种基于表面肌电信号与柔性神经树(Flexible Neural Trees,FNT)模型的实时手势识别模型。表面肌电信号(surface Electromyography,sEMG)具有非入侵式、易于采集特点,故被广泛应用于行为识别和诊断等领域。柔性神经树模型通过简单的预定义来构建,能够解决人工神经网络(Artificial Neural Network,ANN)的结构依赖性高的问题。柔性神经树模型不仅能够避免复杂的计算和电路连接,还具有较高的实时识别率和较低的方均根误差(Root Mean Square Error,RMSE)。实验针对六名参与者的六种手势进行测试,结果表明该模型实时识别率较高,实际应用也证明该算法可行。

Traditional hand gesture recognition methods require designers to use digital signal processing hardware or ensemble methods as tools to solve real time hand gesture classification. These methods are easy to result in com- plicated computational models, complex circuit connection and lower online recognition rate. A real-time hand ges- ture recognition model by combining Flexible Neural Trees （FNT） and surface Electromyography （sEMG） signals is proposed, sEMG is non-invasive, easy to record signal of superficial muscles from the skin surface, which is ap- plied widely in many fields of human motion recognition and diagnosis. The FNT model is generated and evolved based on the pre-defined simple instruction sets, which can solve highly structure dependent problem of the Artifi- cial Neural Network（ANN）. FNT method not only can avoid complicated computation and inconvenience of circuit connection but also has a higher real-time recognition rate and lower Root Mean Square Error（RMSE）. Testing is conducted with six participants to test six hand gestures. The results indicate that the model is a practicable method and has a higher real-time recognition rate.