The Measurement And ANN Prediction Of Human Joint Angle Based On SEMG Recognition

The electromyography (EMG) signal can be considered as a manifestation of themuscle activity,which controls voluntary movements. Surface electromyography(SEMG) has been proved to be a successful method of non-invasive measurement ofEMG and has been often used as a control signal for a prosthetic arm. Most of theresearches inland focus on the mode recognition of SEMG to realize multifunctionalcontrol of artificial limbs. However, in the single motion mode, the factors such asvelocity of movement, amplitude of movement, and position of the arm remainunclear. The relation between SEMG and arm dynamics in the presence of movementhas not been fully understood due to the complexity of EMG generation and thecomplicated relation between EMG and motor control system in the presence ofvoluntary movements.To solve the above problems, a BP artificial neural network to predict the elbow jointangle via SEMG signals is developed in this thesis. This project is sponsored by theNational Natural Science Foundation of China (Project No. 50375108) and theNatural Science Foundation of Tianjin (Project No. 033601611).In the experiment, normal subjects are asked to perform voluntary flexion-extensionmovements on the vertical and horizontal planes respectively. Surface EMG signalsdetected from biceps brachia and triceps brachia are transformed from analog ones todigital ones and put into computer for later processing. The joint angle at the elbow iscalculated from the acceleration of the wrist by a special transformation method. Asignal processing module is developed to analyze the surface EMG signal and extractits instantaneous root mean square (RMS) value. A three-layer BP neural network isconstructed and then is trained by improved back propagation algorism to predict theelbow joint angle by using the RMS of the raw SEMG signal. The network isevaluated based on the best linear regression between the actual joint angle and thepredicted joint angle. The correlation coefficient between these two angles iscalculated for evaluation. Then the predicted angle is used to control the virtual arm ina graph user interface to simulate the real arm.The experimental results show that this neural network model can well represent therelationship between SEMG signals and elbow joint angles. In all 10 test sample pairs,the correlation coefficient between the actual and the predicted joint angles is biggerthan 0.9 on either vertical or horizontal plane. The network accurately predicts variousangles on either plane via SEMG signals. The study has demonstrated a uniqueapproach to determine the movement parameters in a single flexion-extension limbfunction with the aid of neural network.