Identificador e classificador de padrões musculares em pessoas com anomalias no joelho utilizando rede neural Perceptron

Abstract

According to a survey by the World Health Organization, the elderly population is estimated at approximately two billion people by 2050. Falls are a recurring problem for this population. It is estimated that approximately 32% to 42% of people over 70 suffer at least once a year. Laboratory analysis has shown that chronic degenerative diseases, such as knee dysfunctions, are recurrent in the elderly's sensorimotor system. Thus, the work process will be the development of a classifier, using an artificial Perceptron artificial neural network, or which will be responsible for classifying the data collected from an electromyogram located in the patient's lower limbs. For a neural network, tests will be performed to choose how the neurons for the hidden layers. Before feeding the neural network, the signals will be filtered by a fourth-order Butterworth filter in the 0.5 to 500 Hz range for DC signal changes and high frequency electrode noise, then the signals will be segmented in the control windows. 250 ms, and later normalized, for network training an algorithm backpropagation will be used. After a training step, the net will be tested with different data from the training used without you being able to differentiate the muscle pattern of individuals with and without knee dysfunctions (sic).