MYOELECTRIC SIGNAL CLASSIFICATION BASED ON ITS NORMALIZED POWER SPECTRAL DENSITY

摘要

Many studies are realized aiming the control of prosthetic devices using the myoelectric signal acquired on the reminiscent member of the amputee. This work investigates the viability of myoelectric signal recognition by processing and analyzing the power spectral density of this signal. As shown in this text, the proper recognition of four distinct myoelectric signal classes was achieved using this methodology, being each of these classes associated to a different hand motion. Data were acquired using two dry surface electrodes placed on the forearm, and only the normalized power spectral densities of the signals were used during project and evaluation phases. The algorithm presented has been designed to allow real-time evaluation, for practical implementation feasibility. Moreover, since PSD normalization is evaluated before the recognition phase, every absolute value is discarded, allowing the algorithm to ensure some insensibility to impedance fluctuations occurring on the interfaces between the surface electrodes and the skin.