Ανάπτυξη μεθόδων μηχανικής μάθησης (SVMs και NNs) για την πρόβλεψη ιδιοτήτων υλικών με βάση την χρήση δεδομένων ανάστροφου χώρου
Development of machine learning models (SVMs and NNs) for identification of structural properties of materials based on k-Space data
View/
Keywords
Μηχανική μάθηση ; Επεξεργασία σήματος ; k-space ; SVM ; Neural networks ; Support vector machines ; MRI ; X-Ray ; Size distribution ; Ανάστροφος χώρος ; Μαγνητική τομογραφίαAbstract
This thesis investigates the potential application of two new approaches, a Neural Network (NN) model and a Support Vector Machine (SVM) model, for fast estimation of structural properties of materials based on k-space data. The k-space signal for these materials could be acquired methods such as: i) Nuclear Magnetic Resonance Imaging (NMRI) and ii) X-Ray microtomography (X-Ray mCT). In order to train and test our models we used simulated k-space data produced with a numerical method previously developed for a Bayesian prediction technique. Furthermore, using advanced tools available with the Python programming language, we developed a machine learning (ML) pre-processing, training, validation, testing and analysis pipeline. The new models investigated here seem to offer an improved performance compared to existing methods. Finally, suggestions for further work are presented based on the knowledge acquired from this thesis project.