Giorgos Yiannikas

Diploma Work:

System development for user verification and identification using Surface Electromyography (sEMG) signals

Description:

Conducted an advanced research project aimed at developing a robust biometric authentication system utilizing surface electromyography (sEMG) signals. The objective of the system is to allow users to establish a personalized code through distinct hand gestures, which are then captured and stored as unique sEMG signals.

Key Features:

Gesture Enrollment:

• Users register by executing a chosen hand gesture, capturing the corresponding sEMG signal.
• The personalized sEMG signal is stored securely in the system as the user's biometric reference.

Siamese Convolutional Neural Network (CNN):

• Implemented a specialized Siamese CNN architecture for comparative analysis of sEMG signals.
• Trained the network to discern whether two input sEMG signals correspond to the same individual executing the same hand gesture.

Biometric Authentication:

• During authentication, users present their hand gesture, generating a real-time sEMG signal.
• The Siamese CNN evaluates the similarity between the presented signal and the stored biometric reference, allowing or denying access based on a predefined threshold.

Key Achievements:

• Developed a Siamese Convolutional Neural Network (CNN) tailored for processing sEMG signals.
• Integrated data from the Myo armband, a low-cost non-invasive device capable of measuring electromyography signals, as a primary data source.
• Aimed to create a user-friendly application with functionality akin to facial recognition or fingerprint scanning, using sEMG signals for efficient and secure user verification and identification.

Significance:

This project demonstrates a novel approach to biometric authentication using affordable technology, contributing to the advancement of non-invasive user identification systems.

Future Implications:

The research lays the groundwork for the development of practical applications, potentially leading to the creation of secure and cost-effective alternatives to traditional biometric authentication methods.

This diploma work showcases a blend of theoretical knowledge and practical application, highlighting proficiency in signal processing, machine learning, and the integration of emerging technologies for solving real-world challenges.

Academic Projects & Experience

Digital Signal Processing:

• Design of Filters using the Remez algorithm and Linear programming for Min-Max Approximation

Digital Image Processing:

• Image Classification using Convolutional Neural Networks
• Image Compression
• Image Enhancement
• Shape Detection
• Image Restoration
• Image Retrieval from databases

Processing Methods with Machine Learning:

• Bayes Optimal Classifier
• Build from scratch and train Neural Network with loss function a) Hinge, b) Cross-entropy and c) Exponential for binary classification.
• Image restoration when part of it is lost through inpainting or resolution reduction.
• Probability density approximation using Gaussian kernel.
• Application of K-means for clustering

Probabilistic Estimators:

• Kalman Filter

Adaptive Control and Reinforcement Learning:

• Controller Design with MIT rule, MRAC and ADI techniques
• Adaptive control at systems with uncertainties using Neural Networks
• Reinforcement learning control.

Non-linear Control:

• System linearization with input-state method
• System linearization with input-output method

Helping fellow students

In charge of crew for the sound and lighting coverage of public and private events