The application of wavelet transform and neural network to surface electromyographic signals for pattern recognition

The surface electromyographic (SEMG) signal, which is produced by neural and muscular systems, is a complicated bioelectric signal recorded from skin surface using electrodes. It is very helpful for doctors to analysis the illness of patients. In the paper, four channel SEMG signals from four muscles (palmaris longus, brachioradialis, flexor carpi ulnaris, biceps brachii) are analyzed with wavelet transform, and the eigenvalues of 6 layers wavelet decomposition coefficients are distilled, and eigenvector is composed to input the Elman neural network classifier to identify different movement patterns. The eight movement patterns (to make a fist, to spread a fist, wrist circumrotates entad, wrist circumrotates forth, to bend wrist, to spread wrist, forearm circumrotates entad, forearm circumrotates forth) can be successfully identified after training. It is a new method for SEMG signal study and experiments show that it facilitates higher identification rate.