Please use this identifier to cite or link to this item:
|Title:||Κατασκευή συστήματος ηλεκτρομυογραφίας με εφαρμογή σε προσθετικά άκρα|
|Keywords (translated):||Electromyography (EMG)|
|Abstract:||Στο πρώτο κεφάλαιο της πτυχιακής εργασίας αναλύονται η δομή και η λειτουργία του μυός ,καθώς και του νευρικού κυττάρου, με σκοπό την ορθότερη κατανόηση των παραγόμενων ηλεκτρομυικών σημάτων. Επίσης, επεξηγούνται οι όροι “νευρομυική μονάδα” και “νευρομυική σύναψη”. Τέλος , αναλύεται η διαδικασία της νευρικής ώσης και της μυϊκής ενεργοποίησης. Να τονιστεί πως ειδική αναφορά γίνεται στον ανθρώπινο πήχη, καθώς εκεί είναι το σημείο που ‘’διαβάζει’’ ηλεκτρομυικά σήματα και είναι βασιζόμενη η εργασία.|
|Abstract (translated):||In this thesis we study technics of machine learning applied in electromyography based movement recognition. First of all we are building a system which can read, detect and sample surface electromyography signal , through human arm. We are recording signal samples of three movements and based on them we are trying to make a neural network classifier which can recognise these three movements. Each signal we have, we process it in order to eliminate the noise and then we are extracting the features that are more fitting to describe the movement. In this part we tried to extract different features but we found out that the best features are those which are based on the sequences of the signal , extracted with the Fourier Analysis. The neural network we are building in this thesis , created in a specific form code that allows it to change easily the form of the network. That means that the user can choose how many hidden layers the network has , which connections they have with each other and how many cells each hidden layer has. After the building of the network we are trainning it, with supervised methods. We are also trying to do an extensive validation of its performance , not only based on in sample validation. In this section we are also validating the pre-processing we have done , by testing the networks efficiency with raw data. In addition we are trying to make a small simulation , for different kinds of the architecture of our network so we can find the optimal network form. Last but not least we use a new variation of the Shapley Value method to see how each sequence on the signal can feature each movement. In more detail we are trying to see which sequences of the signal, contains the most information for the classification of the movement. We have to also mention , that during this thesis we present theoretically basics algorithms and technics of the machine learning field that are commonly applied. This is knowledge , that even though has not applied in this thesis, every engineer should have.|
|Appears in Collections:||Τμήμα Ηλεκτρολ. Μηχαν. και Τεχνολ. Υπολογ. (ΔΕ)|
Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.