Development of an embedded system for classification of EMG signals

Abstract

This article describes a methodology to create high-level algorithms, such as artificial neural networks and other pattern recognition techniques to be implemented in embedded systems. A system was developed on a DSP (Digital Signal Processor), for identification and classification of EMG signals; for this purpose the Code Composer Studio software V3.3 and Matlab package were coupled for programming the TMS320F28335 card. The aim of this study was to create a model of an artificial neural network from a previous pre-processing to be implemented on an embedded hardware with the respective network. A typical MLP (Multilayer Perceptron) Network was developed from a pre-processed data set, which were obtained through the acquisition of EMG signals. The data were validated with the discrimination technique as Principal Components Analysis (PCA), which was useful to determine the repeatability and selectivity of the measuring system. Through this application it was possible to improve the processing speed, portability and response of EMG device, which opens a wide range of possibilities for this methodology to be applied in different sectors (e.g, industry, health, etc.) and mainly as a signal classification system.